![[BACK]](/cvsweb/icons/back.gif){kind=icon}
Return to internals.texinfo CVS log ![[TXT]](/cvsweb/icons/text.gif){kind=icon}
![[DIR]](/cvsweb/icons/dir.gif){kind=icon}
Up to [GCC 1.x] / gcc|
Return to internals.texinfo CVS log | Up to [GCC 1.x] / gcc |
1.1 root 1: \input texinfo @c -*-texinfo-*-
2:
3: @settitle Internals of GNU CC
4: @setfilename internals
5:
6: @ifinfo
7: This file documents the internals of the GNU compiler.
8:
1.1.1.2 root 9: Copyright (C) 1988 Free Software Foundation, Inc.
1.1 root 10:
11: Permission is granted to make and distribute verbatim copies of
12: this manual provided the copyright notice and this permission notice
13: are preserved on all copies.
14:
15: @ignore
16: Permission is granted to process this file through Tex and print the
17: results, provided the printed document carries copying permission
18: notice identical to this one except for the removal of this paragraph
19: (this paragraph not being relevant to the printed manual).
20:
21: @end ignore
22: Permission is granted to copy and distribute modified versions of this
23: manual under the conditions for verbatim copying, provided also that the
24: section entitled ``GNU CC General Public License'' is included exactly as
25: in the original, and provided that the entire resulting derived work is
26: distributed under the terms of a permission notice identical to this one.
27:
28: Permission is granted to copy and distribute translations of this manual
29: into another language, under the above conditions for modified versions,
1.1.1.2 root 30: except that the section entitled ``GNU CC General Public License'' and
31: this permission notice may be included in translations approved by the
32: Free Software Foundation instead of in the original English.
1.1 root 33: @end ifinfo
34:
35: @setchapternewpage odd
36:
37: @titlepage
38: @center @titlefont{Internals of GNU CC}
39: @sp 2
40: @center Richard M. Stallman
1.1.1.2 root 41: @sp 3
1.1.1.4 root 42: @center last updated 26 June 1988
1.1.1.2 root 43: @sp 1
1.1.1.4 root 44: @center for version 1.23
1.1 root 45: @page
46: @vskip 0pt plus 1filll
1.1.1.2 root 47: Copyright @copyright{} 1988 Free Software Foundation, Inc.
1.1 root 48:
49: Permission is granted to make and distribute verbatim copies of
50: this manual provided the copyright notice and this permission notice
51: are preserved on all copies.
52:
53: Permission is granted to copy and distribute modified versions of this
54: manual under the conditions for verbatim copying, provided also that the
55: section entitled ``GNU CC General Public License'' is included exactly as
56: in the original, and provided that the entire resulting derived work is
57: distributed under the terms of a permission notice identical to this one.
58:
59: Permission is granted to copy and distribute translations of this manual
60: into another language, under the above conditions for modified versions,
61: except that the section entitled ``GNU CC General Public License'' may be
62: included in a translation approved by the author instead of in the original
63: English.
64: @end titlepage
65: @page
66:
67: @ifinfo
1.1.1.2 root 68: @node Top, Copying,, (DIR)
69: @ichapter Introduction
1.1 root 70:
1.1.1.2 root 71: This manual documents how to run, install and port the GNU C compiler, as
72: well as its new features and incompatibilities, and how to report bugs.
1.1 root 73:
74: @end ifinfo
75: @menu
76: * Copying:: GNU CC General Public License says
77: how you can copy and share GNU CC.
1.1.1.2 root 78: * Contributors:: People who have contributed to GNU CC.
79: * Options:: Command options supported by @samp{gcc}.
1.1 root 80: * Installation:: How to configure, compile and install GNU CC.
1.1.1.3 root 81: * Trouble:: If you have trouble installing GNU CC.
1.1.1.2 root 82: * Incompatibilities:: Incompatibilities of GNU CC.
83: * Extensions:: GNU extensions to the C language.
84: * Bugs:: How to report bugs (if you want to get them fixed).
1.1 root 85: * Portability:: Goals of GNU CC's portability features.
1.1.1.2 root 86: * Interface:: Function-call interface of GNU CC output.
1.1 root 87: * Passes:: Order of passes, what they do, and what each file is for.
88: * RTL:: The intermediate representation that most passes work on.
89: * Machine Desc:: How to write machine description instruction patterns.
90: * Machine Macros:: How to write the machine description C macros.
91: @end menu
92:
1.1.1.2 root 93: @node Copying, Contributors, Top, Top
1.1 root 94: @unnumbered GNU CC GENERAL PUBLIC LICENSE
1.1.1.2 root 95: @center (Clarified 11 Feb 1988)
1.1 root 96:
97: The license agreements of most software companies keep you at the
98: mercy of those companies. By contrast, our general public license is
99: intended to give everyone the right to share GNU CC. To make sure that
100: you get the rights we want you to have, we need to make restrictions
101: that forbid anyone to deny you these rights or to ask you to surrender
102: the rights. Hence this license agreement.
103:
1.1.1.2 root 104: Specifically, we want to make sure that you have the right to give
105: away copies of GNU CC, that you receive source code or else can get it
106: if you want it, that you can change GNU CC or use pieces of it in new
107: free programs, and that you know you can do these things.
108:
109: To make sure that everyone has such rights, we have to forbid you to
110: deprive anyone else of these rights. For example, if you distribute
111: copies of GNU CC, you must give the recipients all the rights that you
112: have. You must make sure that they, too, receive or can get the
113: source code. And you must tell them their rights.
114:
115: Also, for our own protection, we must make certain that everyone
116: finds out that there is no warranty for GNU CC. If GNU CC is modified by
117: someone else and passed on, we want its recipients to know that what
118: they have is not what we distributed, so that any problems introduced
119: by others will not reflect on our reputation.
120:
121: Therefore we (Richard Stallman and the Free Software Foundation,
122: Inc.) make the following terms which say what you must do to be
123: allowed to distribute or change GNU CC.
124:
125: @unnumberedsec COPYING POLICIES
126:
127: @enumerate
128: @item
129: You may copy and distribute verbatim copies of GNU CC source code as
130: you receive it, in any medium, provided that you conspicuously and
131: appropriately publish on each copy a valid copyright notice
132: ``Copyright @copyright{} 1988 Free Software Foundation, Inc.'' (or
133: with whatever year is appropriate); keep intact the notices on all
134: files that refer to this License Agreement and to the absence of any
135: warranty; and give any other recipients of the GNU CC program a copy
136: of this License Agreement along with the program. You may charge a
137: distribution fee for the physical act of transferring a copy.
138:
139: @item
140: You may modify your copy or copies of GNU CC or any portion of it,
141: and copy and distribute such modifications under the terms of
142: Paragraph 1 above, provided that you also do the following:
143:
144: @itemize @bullet
145: @item
146: cause the modified files to carry prominent notices stating
147: that you changed the files and the date of any change; and
148:
149: @item
150: cause the whole of any work that you distribute or publish, that
151: in whole or in part contains or is a derivative of GNU CC or any
152: part thereof, to be licensed at no charge to all third parties on
153: terms identical to those contained in this License Agreement
154: (except that you may choose to grant more extensive warranty
155: protection to some or all third parties, at your option).
156:
157: @item
158: You may charge a distribution fee for the physical act of
159: transferring a copy, and you may at your option offer warranty
160: protection in exchange for a fee.
161: @end itemize
162:
163: Mere aggregation of another unrelated program with this program (or its
164: derivative) on a volume of a storage or distribution medium does not bring
165: the other program under the scope of these terms.
166:
167: @item
168: You may copy and distribute GNU CC (or a portion or derivative of it,
169: under Paragraph 2) in object code or executable form under the terms
170: of Paragraphs 1 and 2 above provided that you also do one of the
171: following:
172:
173: @itemize @bullet
174: @item
175: accompany it with the complete corresponding machine-readable
176: source code, which must be distributed under the terms of
177: Paragraphs 1 and 2 above; or,
178:
179: @item
180: accompany it with a written offer, valid for at least three
181: years, to give any third party free (except for a nominal
182: shipping charge) a complete machine-readable copy of the
183: corresponding source code, to be distributed under the terms of
184: Paragraphs 1 and 2 above; or,
185:
186: @item
187: accompany it with the information you received as to where the
188: corresponding source code may be obtained. (This alternative is
189: allowed only for noncommercial distribution and only if you
190: received the program in object code or executable form alone.)
191: @end itemize
192:
193: For an executable file, complete source code means all the source code
194: for all modules it contains; but, as a special exception, it need not
195: include source code for modules which are standard libraries that
196: accompany the operating system on which the executable file runs.
197:
198: @item
199: You may not copy, sublicense, distribute or transfer GNU CC except as
200: expressly provided under this License Agreement. Any attempt
201: otherwise to copy, sublicense, distribute or transfer GNU CC is void
202: and your rights to use the program under this License agreement shall
203: be automatically terminated. However, parties who have received
204: computer software programs from you with this License Agreement will
205: not have their licenses terminated so long as such parties remain in
206: full compliance.
207:
208: @item
209: If you wish to incorporate parts of GNU CC into other free programs
210: whose distribution conditions are different, write to the Free Software
211: Foundation at 675 Mass Ave, Cambridge, MA 02139. We have not yet worked
212: out a simple rule that can be stated here, but we will often permit this.
213: We will be guided by the two goals of preserving the free status of all
214: derivatives of our free software and of promoting the sharing and reuse of
215: software.
216: @end enumerate
217:
218: Your comments and suggestions about our licensing policies and our
219: software are welcome! Please contact the Free Software Foundation, Inc.,
220: 675 Mass Ave, Cambridge, MA 02139, or call (617) 876-3296.
221:
222: @unnumberedsec NO WARRANTY
223:
224: BECAUSE GNU CC IS LICENSED FREE OF CHARGE, WE PROVIDE ABSOLUTELY NO
225: WARRANTY, TO THE EXTENT PERMITTED BY APPLICABLE STATE LAW. EXCEPT
226: WHEN OTHERWISE STATED IN WRITING, FREE SOFTWARE FOUNDATION, INC,
227: RICHARD M. STALLMAN AND/OR OTHER PARTIES PROVIDE GNU CC "AS IS" WITHOUT
228: WARRANTY OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT
229: LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
230: A PARTICULAR PURPOSE. THE ENTIRE RISK AS TO THE QUALITY AND
231: PERFORMANCE OF GNU CC IS WITH YOU. SHOULD GNU CC PROVE DEFECTIVE, YOU
232: ASSUME THE COST OF ALL NECESSARY SERVICING, REPAIR OR CORRECTION.
233:
234: IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW WILL RICHARD M.
235: STALLMAN, THE FREE SOFTWARE FOUNDATION, INC., AND/OR ANY OTHER PARTY
236: WHO MAY MODIFY AND REDISTRIBUTE GNU CC AS PERMITTED ABOVE, BE LIABLE TO
237: YOU FOR DAMAGES, INCLUDING ANY LOST PROFITS, LOST MONIES, OR OTHER
238: SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE USE OR
239: INABILITY TO USE (INCLUDING BUT NOT LIMITED TO LOSS OF DATA OR DATA
240: BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY THIRD PARTIES OR A
241: FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER PROGRAMS) GNU CC, EVEN
242: IF YOU HAVE BEEN ADVISED OF THE POSSIBILITY OF SUCH DAMAGES, OR FOR
243: ANY CLAIM BY ANY OTHER PARTY.
244:
245: @node Contributors, Options, Copying, Top
246: @unnumbered Contributors to GNU CC
247:
248: In addition to Richard Stallman, several people have written parts
249: of GNU CC.
250:
251: @itemize @bullet
252: @item
253: The idea of using RTL and some of the optimization ideas came from the
254: U. of Arizona Portable Optimizer, written by Jack Davidson and
255: Christopher Fraser. See ``Register Allocation and Exhaustive Peephole
256: Optimization'', Software Practice and Experience 14 (9), Sept. 1984,
257: 857-866.
258:
259: @item
260: Paul Rubin wrote most of the preprocessor.
261:
262: @item
263: Leonard Tower wrote parts of the parser, RTL generator, RTL
264: definitions, and of the Vax machine description.
265:
266: @item
267: Ted Lemon wrote parts of the RTL reader and printer.
268:
269: @item
270: Nobuyuki Hikichi of Software Research Associates, Tokyo, contributed
271: the support for the SONY NEWS machine.
272:
273: @item
274: Charles LaBrec contributed the support for the Integrated Solutions
275: 68020 system.
276:
277: @item
278: Michael Tiemann of MCC wrote the description of the National
279: Semiconductor 32000 series cpu, with some contributions from Jan Stein
280: of the Chalmers Computer Club. Tiemann also wrote the code for inline
1.1.1.4 root 281: function integration and for the SPARC cpu.
1.1.1.2 root 282:
283: @item
1.1.1.3 root 284: Robert Brown implemented the support for Encore 32000 systems.
285:
286: @item
1.1.1.4 root 287: David Kashtan of SRI adapted GNU CC to the Vomit-Making System.
1.1.1.2 root 288:
289: @item
290: Alex Crain provided changes for the 3b1.
291:
292: @item
1.1.1.4 root 293: Greg Satz and Chris Hanson assisted in making GNU CC work on HP-UX for
294: the 9000 series 300.
1.1.1.2 root 295: @end itemize
296:
297: @node Options, Installation, Contributors, Top
298: @chapter GNU CC Command Options
299:
300: The GNU C compiler uses a command syntax much like the Unix C compiler.
301: The @code{gcc} program accepts options and file names as operands.
302: Multiple single-letter options may @emph{not} be grouped: @samp{-dr} is
303: very different from @samp{-d -r}.
304:
305: When you invoke GNU CC, it normally does preprocessing, compilation,
306: assembly and linking. File names which end in @samp{.c} are taken as C
307: source to be preprocessed and compiled; compiler output files plus any
308: input files with names ending in @samp{.s} are assembled; then the
309: resulting object files, plus any other input files, are linked together to
310: produce an executable.
311:
312: Command options allow you to stop this process at an intermediate stage.
313: For example, the @samp{-c} option says not to run the linker. Then the
314: output consists of object files output by the assembler.
315:
316: Other command options are passed on to one stage. Some options control
317: the preprocessor and others the compiler itself. Yet other options
318: control the assembler and linker; these are not documented here because the
319: GNU assembler and linker are not yet released.
320:
321: Here are the options to control the overall compilation process, including
322: those that say whether to link, whether to assemble, and so on.
323:
324: @table @samp
325: @item -o @var{file}
326: Place output in file @var{file}. This applies regardless to whatever
327: sort of output is being produced, whether it be an executable file,
328: an object file, an assembler file or preprocessed C code.
329:
1.1.1.3 root 330: If @samp{-o} is not specified, the default is to put an executable file
1.1.1.2 root 331: in @file{a.out}, the object file @file{@var{source}.c} in
332: @file{@var{source}.o}, an assembler file in @file{@var{source}.s}, and
333: preprocessed C on standard output.@refill
334:
335: @item -c
336: Compile or assemble the source files, but do not link. Produce object
337: files with names made by replacing @samp{.c} or @samp{.s} with
338: @samp{.o} at the end of the input file names. Do nothing at all for
339: object files specified as input.
340:
341: @item -S
342: Compile into assembler code but do not assemble. The assembler output
343: file name is made by replacing @samp{.c} with @samp{.s} at the end of
344: the input file name. Do nothing at all for assembler source files or
345: object files specified as input.
346:
347: @item -E
348: Run only the C preprocessor. Preprocess all the C source files
349: specified and output the results to standard output.
350:
351: @item -v
352: Compiler driver program prints the commands it executes as it runs
353: the preprocessor, compiler proper, assembler and linker. Some of
354: these are directed to print their own version numbers.
355:
356: @item -B@var{prefix}
357: Compiler driver program tries @var{prefix} as a prefix for each
358: program it tries to run. These programs are @file{cpp}, @file{cc1},
359: @file{as} and @file{ld}.
360:
361: For each subprogram to be run, the compiler driver first tries the
362: @samp{-B} prefix, if any. If that name is not found, or if @samp{-B}
363: was not specified, the driver tries two standard prefixes, which are
364: @file{/usr/lib/gcc-} and @file{/usr/local/lib/gcc-}. If neither of
365: those results in a file name that is found, the unmodified program
366: name is searched for using the directories specified in your
367: @samp{PATH} environment variable.
368:
369: The run-time support file @file{gnulib} is also searched for using
370: the @samp{-B} prefix, if needed. If it is not found there, the two
371: standard prefixes above are tried, and that is all. The file is left
372: out of the link if it is not found by those means. Most of the time,
373: on most machines, you can do without it.
374: @end table
375:
376: These options control the details of C compilation itself.
377:
378: @table @samp
379: @item -ansi
380: Support all ANSI standard C programs.
381:
382: This turns off certain features of GNU C that are incompatible with
383: ANSI C, such as the @code{asm}, @code{inline} and @code{typeof}
384: keywords, and predefined macros such as @code{unix} and @code{vax}
385: that identify the type of system you are using. It also enables the
386: undesirable and rarely used ANSI trigraph feature.
387:
388: The @samp{-ansi} option does not cause non-ANSI programs to be
389: rejected gratuitously. For that, @samp{-pedantic} is required in
390: addition to @samp{-ansi}.
391:
392: The macro @code{__STRICT_ANSI__} is predefined when the @samp{-ansi}
393: option is used. Some header files may notice this macro and refrain
394: from declaring certain functions or defining certain macros that the
395: ANSI standard doesn't call for; this is to avoid interfering with
396: any programs that might use these names for other things.
397:
398: @item -traditional
399: Attempt to support some aspects of traditional C compilers.
400: Specifically:
401:
402: @itemize @bullet
403: @item
404: All @code{extern} declarations take effect globally even if they
405: are written inside of a function definition. This includes implicit
406: declarations of functions.
407:
408: @item
409: The keywords @code{typeof}, @code{inline}, @code{signed}, @code{const}
410: and @code{volatile} are not recognized.@refill
411:
412: @item
413: Comparisons between pointers and integers are always allowed.
414:
415: @item
416: Integer types @code{unsigned short} and @code{unsigned char} promote
417: to @code{unsigned int}.
418:
419: @item
1.1.1.5 ! root 420: Out-of-range floating point literals are not an error.
! 421:
! 422: @item
1.1.1.2 root 423: In the preprocessor, comments convert to nothing at all, rather than to
424: a space. This allows traditional token concatenation.
425:
426: @item
427: In the preprocessor, single and double quote characters are ignored
428: when scanning macro definitions, so that macro arguments can be replaced
429: even within a string or character constant. Quote characters are also
430: ignored when skipping text inside a failing conditional directive.
431: @end itemize
432:
433: @item -O
434: Optimize. Optimizing compilation takes somewhat more time, and a lot
435: more memory for a large function.
436:
437: Without @samp{-O}, the compiler's goal is to reduce the cost of
438: compilation and to make debugging produce the expected results.
439: Statements are independent: if you stop the program with a breakpoint
440: between statements, you can then assign a new value to any variable or
441: change the program counter to any other statement in the function and
442: get exactly the results you would expect from the source code.
443:
444: Without @samp{-O}, only variables declared @code{register} are
445: allocated in registers. The resulting compiled code is a little worse
446: than produced by PCC without @samp{-O}.
447:
448: With @samp{-O}, the compiler tries to reduce code size and execution
449: time.
450:
451: Some of the @samp{-f} options described below turn specific kinds of
452: optimization on or off.
453:
454: @item -g
1.1.1.4 root 455: Produce debugging information in the operating system's native
456: format (for DBX or SDB).
1.1.1.2 root 457:
458: Unlike most other C compilers, GNU CC allows you to use @samp{-g} with
459: @samp{-O}. The shortcuts taken by optimized code may occasionally
460: produce surprising results: some variables you declared may not exist
461: at all; flow of control may briefly move where you did not expect it;
462: some statements may not be executed because they compute constant
463: results or their values were already at hand; some statements may
464: execute in different places because they were moved out of loops.
465: Nevertheless it proves possible to debug optimized output. This makes
466: it reasonable to use the optimizer for programs that might have bugs.
467:
468: @item -gg
469: Produce debugging information in GDB's own format. This requires
470: the GNU assembler and linker in order to work.
471:
472: @item -w
473: Inhibit all warning messages.
474:
475: @item -W
476: Print extra warning messages for these events:
477:
478: @itemize @bullet
479: @item
480: An automatic variable is used without first being initialized.
481:
482: These warnings are possible only in optimizing compilation,
483: because they require data flow information that is computed only
484: when optimizing. They occur only for variables that are
485: candidates for register allocation. Therefore, they do not occur
486: for a variable that is declared @code{volatile}, or whose address
487: is taken, or whose size is other than 1, 2, 4 or 8 bytes. Also,
488: they do not occur for structures, unions or arrays, even when
489: they are in registers.
490:
491: Note that there may be no warning about a variable that is used
492: only to compute a value that itself is never used, because such
493: computations may be deleted by the flow analysis pass before the
494: warnings are printed.
495:
496: These warnings are made optional because GNU CC is not smart
497: enough to see all the reasons why the code might be correct
498: despite appearing to have an error. Here is one example of how
499: this can happen:
500:
501: @example
502: @{
503: int x;
504: switch (y)
505: @{
506: case 1: x = 1;
507: break;
508: case 2: x = 4;
509: break;
510: case 3: x = 5;
511: @}
512: foo (x);
513: @}
514: @end example
515:
516: @noindent
517: If the value of @code{y} is always 1, 2 or 3, then @code{x} is
518: always initialized, but GNU CC doesn't know this. Here is
519: another common case:
520:
521: @example
522: @{
523: int save_y;
524: if (change_y) save_y = y, y = new_y;
525: @dots{}
526: if (change_y) y = save_y;
527: @}
528: @end example
529:
530: @noindent
1.1.1.4 root 531: This has no bug because @code{save_y} is used only if it is set.
1.1.1.2 root 532:
533: @item
534: A nonvolatile automatic variable might be changed by a call to
535: @code{longjmp}. These warnings as well are possible only in
536: optimizing compilation.
537:
538: The compiler sees only the calls to @code{setjmp}. It cannot know
539: where @code{longjmp} will be called; in fact, a signal handler could
540: call it at any point in the code. As a result, you may get a warning
541: even when there is in fact no problem because @code{longjmp} cannot
542: in fact be called at the place which would cause a problem.
543:
544: @item
545: A function can return either with or without a value. (Falling
546: off the end of the function body is considered returning without
547: a value.) For example, this function would inspire such a
548: warning:
549:
550: @example
551: foo (a)
552: @{
553: if (a > 0)
554: return a;
555: @}
556: @end example
557:
558: Spurious warnings can occur because GNU CC does not realize that
559: certain functions (including @code{abort} and @code{longjmp})
560: will never return.
561: @end itemize
562:
563: In the future, other useful warnings may also be enabled by this
564: option.
565:
566: @item -Wimplicit
567: Warn whenever a function is implicitly declared.
568:
569: @item -Wreturn-type
570: Warn whenever a function is defined with a return-type that defaults
571: to @code{int}. Also warn about any @code{return} statement with no
572: return-value in a function whose return-type is not @code{void}.
573:
574: @item -Wcomment
575: Warn whenever a comment-start sequence @samp{/*} appears in a comment.
576:
1.1.1.3 root 577: @item -Wall
578: All of the above @samp{-W} options combined.
579:
1.1.1.2 root 580: @item -p
581: Generate extra code to write profile information suitable for the
582: analysis program @code{prof}.
583:
584: @item -pg
585: Generate extra code to write profile information suitable for the
586: analysis program @code{gprof}.
587:
1.1.1.3 root 588: @item -l@var{library}
589: Search a standard list of directories for a library named
590: @var{library}, which is actually a file named
591: @file{lib@var{library}.a}. The linker uses this file as if it
592: had been specified precisely by name.
593:
594: The directories searched include several standard system directories
595: plus any that you specify with @samp{-L}.
596:
597: Normally the files found this way are library files---archive files
598: whose members are object files. The linker handles an archive file by
1.1.1.4 root 599: scanning through it for members which define symbols that have so far
600: been referenced but not defined. But if the file that is found is an
1.1.1.3 root 601: ordinary object file, it is linked in the usual fashion. The only
1.1.1.4 root 602: difference between using an @samp{-l} option and specifying a file name
603: is that @samp{-l} searches several directories.
1.1.1.3 root 604:
605: @item -L@var{dir}
606: Add directory @var{dir} to the list of directories to be searched
607: for @samp{-l}.
1.1.1.2 root 608:
609: @item -nostdlib
610: Don't use the standard system libraries and startup files when
611: linking. Only the files you specify (plus @file{gnulib}) will be
612: passed to the linker.
613:
614: @item -m@var{machinespec}
615: Machine-dependent option specifying something about the type of target
616: machine. These options are defined by the macro
617: @code{TARGET_SWITCHES} in the machine description. The default for
618: the options is also defined by that macro, which enables you to change
619: the defaults.@refill
620:
621: These are the @samp{-m} options defined in the 68000 machine
622: description:
623:
624: @table @samp
625: @item -m68020
626: Generate output for a 68020 (rather than a 68000). This is the
627: default if you use the unmodified sources.
628:
629: @item -m68000
630: Generate output for a 68000 (rather than a 68020).
631:
632: @item -m68881
633: Generate output containing 68881 instructions for floating point.
634: This is the default if you use the unmodified sources.
635:
636: @item -msoft-float
637: Generate output containing library calls for floating point.
638:
639: @item -mshort
640: Consider type @code{int} to be 16 bits wide, like @code{short int}.
641:
642: @item -mnobitfield
643: Do not use the bit-field instructions. @samp{-m68000} implies
644: @samp{-mnobitfield}.
645:
646: @item -mbitfield
647: Do use the bit-field instructions. @samp{-m68020} implies
648: @samp{-mbitfield}. This is the default if you use the unmodified
649: sources.
650:
651: @item -mrtd
652: Use a different function-calling convention, in which functions
653: that take a fixed number of arguments return with the @code{rtd}
654: instruction, which pops their arguments while returning. This
655: saves one instruction in the caller since there is no need to pop
656: the arguments there.
657:
658: This calling convention is incompatible with the one normally
659: used on Unix, so you cannot use it if you need to call libraries
660: compiled with the Unix compiler.
661:
662: Also, you must provide function prototypes for all functions that
663: take variable numbers of arguments (including @code{printf});
664: otherwise incorrect code will be generated for calls to those
665: functions.
666:
667: In addition, seriously incorrect code will result if you call a
668: function with too many arguments. (Normally, extra arguments are
669: harmlessly ignored.)
670:
671: The @code{rtd} instruction is supported by the 68010 and 68020
672: processors, but not by the 68000.
673: @end table
674:
675: These @samp{-m} options are defined in the Vax machine description:
676:
677: @table @samp
678: @item -munix
679: Do not output certain jump instructions (@code{aobleq} and so on)
680: that the Unix assembler for the Vax cannot handle across long
681: ranges.
682:
683: @item -mgnu
684: Do output those jump instructions, on the assumption that you
685: will assemble with the GNU assembler.
1.1.1.3 root 686:
687: @item -mg
688: Output code for g-format floating point numbers instead of d-format.
1.1.1.2 root 689: @end table
690:
691: @item -f@var{flag}
692: Specify machine-independent flags. These are the flags:
693:
694: @table @samp
695: @item -ffloat-store
696: Do not store floating-point variables in registers. This
697: prevents undesirable excess precision on machines such as the
698: 68000 where the floating registers (of the 68881) keep more
699: precision than a @code{double} is supposed to have.
700:
701: For most programs, the excess precision does only good, but a few
702: programs rely on the precise definition of IEEE floating point.
703: Use @samp{-ffloat-store} for such programs.
704:
705: @item -fno-asm
706: Do not recognize @code{asm}, @code{inline} or @code{typeof} as a
707: keyword. These words may then be used as identifiers.
708:
709: @item -fno-defer-pop
710: Always pop the arguments to each function call as soon as that
711: function returns. Normally the compiler (when optimizing) lets
712: arguments accumulate on the stack for several function calls and
713: pops them all at once.
714:
715: @item -fcombine-regs
716: Allow the combine pass to combine an instruction that copies one
717: register into another. This might or might not produce better
718: code when used in addition to @samp{-O}. I am interested in
719: hearing about the difference this makes.
720:
721: @item -fforce-mem
722: Force memory operands to be copied into registers before doing
723: arithmetic on them. This may produce better code by making all
724: memory references potential common subexpressions. When they are
725: not common subexpressions, instruction combination should
726: eliminate the separate register-load. I am interested in hearing
727: about the difference this makes.
728:
729: @item -fforce-addr
730: Force memory address constants to be copied into registers before
731: doing arithmetic on them. This may produce better code just as
732: @samp{-fforce-mem} may. I am interested in hearing about the
733: difference this makes.
734:
735: @item -fomit-frame-pointer
736: Don't keep the frame pointer in a register for functions that
737: don't need one. This avoids the instructions to save, set up and
738: restore frame pointers; it also makes an extra register available
739: in many functions. @strong{It also makes debugging impossible.}
740:
741: On some machines, such as the Vax, this flag has no effect,
742: because the standard calling sequence automatically handles the
743: frame pointer and nothing is saved by pretending it doesn't
744: exist. The machine-description macro
745: @code{FRAME_POINTER_REQUIRED} controls whether a target machine
746: supports this flag. @xref{Registers}.@refill
747:
748: @item -finline-functions
749: Integrate all simple functions into their callers. The compiler
750: heuristically decides which functions are simple enough to be
751: worth integrating in this way.
752:
753: If all calls to a given function are integrated, and the function
754: is declared @code{static}, then the function is normally not
755: output as assembler code in its own right.
756:
757: @item -fkeep-inline-functions
758: Even if all calls to a given function are integrated, and the
759: function is declared @code{static}, nevertheless output a
760: separate run-time callable version of the function.
761:
762: @item -fwritable-strings
763: Store string constants in the writable data segment and don't
764: uniquize them. This is for compatibility with old programs which
765: assume they can write into string constants. Writing into string
766: constants is a very bad idea; ``constants'' should be constant.
767:
768: @item -fno-function-cse
769: Do not put function addresses in registers; make each instruction
770: that calls a constant function contain the function's address
771: explicitly.
772:
773: This option results in less efficient code, but some strange
774: hacks that alter the assembler output may be confused by the
775: optimizations performed when this option is not used.
776:
777: @item -fvolatile
778: Consider all memory references through pointers to be volatile.
779:
780: @item -funsigned-char
781: Let the type @code{char} be the unsigned, like @code{unsigned
782: char}.
783:
784: Each kind of machine has a default for what @code{char} should
785: be. It is either like @code{unsigned char} by default or like
786: @code{signed char} by default. (Actually, at present, the
787: default is always signed.)
788:
789: The type @code{char} is always a distinct type from either
790: @code{signed char} or @code{unsigned char}, even though its
791: behavior is always just like one of those two.
792:
793: @item -fsigned-char
794: Let the type @code{char} be signed, like @code{signed char}.
795:
796: @item -ffixed-@var{reg}
797: Treat the register named @var{reg} as a fixed register; generated
798: code should never refer to it (except perhaps as a stack pointer,
799: frame pointer or in some other fixed role).
800:
801: @var{reg} must be the name of a register. The register names
802: accepted are machine-specific and are defined in the
803: @code{REGISTER_NAMES} macro in the machine description macro
804: file.
805:
806: @item -fcall-used-@var{reg}
807: Treat the register named @var{reg} as an allocatable register
808: that is clobbered by function calls. It may be allocated for
809: temporaries or variables that do not live across a call.
810: Functions compiled this way will not save and restore the
811: register @var{reg}.
812:
813: Use of this flag for a register that has a fixed pervasive role
814: in the machine's execution model, such as the stack pointer or
815: frame pointer, will produce disastrous results.
816:
817: @item -fcall-saved-@var{reg}
818: Treat the register named @var{reg} as an allocatable register
819: saved by functions. It may be allocated even for temporaries or
820: variables that live across a call. Functions compiled this way
821: will save and restore the register @var{reg} if they use it.
822:
823: Use of this flag for a register that has a fixed pervasive role
824: in the machine's execution model, such as the stack pointer or
825: frame pointer, will produce disastrous results.
826:
827: A different sort of disaster will result from the use of this
828: flag for a register in which function values are may be returned.
829: @end table
830:
831: @item -d@var{letters}
832: Says to make debugging dumps at times specified by @var{letters}.
833: Here are the possible letters:
834:
835: @table @samp
836: @item r
837: Dump after RTL generation.
838: @item j
839: Dump after first jump optimization.
840: @item J
841: Dump after last jump optimization.
842: @item s
843: Dump after CSE (including the jump optimization that sometimes
844: follows CSE).
845: @item L
846: Dump after loop optimization.
847: @item f
848: Dump after flow analysis.
849: @item c
850: Dump after instruction combination.
851: @item l
852: Dump after local register allocation.
853: @item g
854: Dump after global register allocation.
855: @item m
856: Print statistics on memory usage, at the end of the run.
857: @end table
858:
859: @item -pedantic
860: Issue all the warnings demanded by strict ANSI standard C; reject
861: all programs that use forbidden extensions.
862:
863: Valid ANSI standard C programs should compile properly with or without
864: this option (though a rare few will require @samp{-ansi}). However,
865: without this option, certain GNU extensions and traditional C features
866: are supported as well. With this option, they are rejected. There is
867: no reason to @i{use} this option; it exists only to satisfy pedants.
868: @end table
869:
870: These options control the C preprocessor, which is run on each C source
871: file before actual compilation. If you use the @samp{-E} option, nothing
872: is done except C preprocessing. Some of these options make sense only
873: together with @samp{-E} because they request preprocessor output that is
874: not suitable for actual compilation.
875:
876: @table @samp
877: @item -C
878: Tell the preprocessor not to discard comments. Used with the
879: @samp{-E} option.
880:
881: @item -I@var{dir}
882: Search directory @var{dir} for include files.
883:
1.1.1.3 root 884: @item -I-
885: Any directories specified with @samp{-I} options before the @samp{-I-}
886: option are searched only for the case of @samp{#include "@var{file}"};
887: they are not searched for @samp{#include <@var{file}>}.
888:
889: If additional directories are specified with @samp{-I} options after
890: the @samp{-I-}, these directories are searched for all @samp{#include}
891: directives. (Ordinarily @emph{all} @samp{-I} directories are used
892: this way.)
893:
894: In addition, the @samp{-I-} option inhibits the use of the current
895: directory as the first search directory for @samp{#include
896: "@var{file}"}. Therefore, the current directory is searched only if
897: it is requested explicitly with @samp{-I.}. Specifying both
898: @samp{-I-} and @samp{-I.} allows you to control precisely which
899: directories are searched before the current one and which are searched
900: after.
901:
902: @item -nostdinc
903: Do not search the standard system directories for header files. Only
904: the directories you have specified with @samp{-I} options (and the
905: current directory, if appropriate) are searched.
906:
907: Between @samp{-nostdinc} and @samp{-I-}, you can eliminate all
908: directories from the search path except those you specify.
909:
1.1.1.2 root 910: @item -M
911: Tell the preprocessor to output a rule suitable for @code{make}
912: describing the dependencies of each source file. For each source
913: file, the preprocessor outputs one @code{make}-rule whose target is
914: the object file name for that source file and whose dependencies are
915: all the files @samp{#include}d in it. This rule may be a single line
916: or may be continued with @samp{\}-newline if it is long.
917:
918: @samp{-M} implies @samp{-E}.
919:
920: @item -MM
921: Like @samp{-M} but the output mentions only the user-header files
922: included with @samp{#include "@var{file}"}. System header files
923: included with @samp{#include <@var{file}>} are omitted.
924:
925: @samp{-MM} implies @samp{-E}.
926:
927: @item -D@var{macro}
928: Define macro @var{macro} with the empty string as its definition.
929:
930: @item -D@var{macro}=@var{defn}
931: Define macro @var{macro} as @var{defn}.
932:
933: @item -U@var{macro}
934: Undefine macro @var{macro}.
935:
936: @item -T
937: Support ANSI C trigraphs. You don't want to know about this
938: brain-damage. The @samp{-ansi} option also has this effect.
939: @end table
940:
1.1.1.3 root 941: @node Installation, Trouble, Options, Top
1.1.1.2 root 942: @chapter Installing GNU CC
943:
944: Here is the procedure for installing GNU CC on a Unix system.
945: @menu
946: * VMS Install:: See below for installation on VMS.
947: @end menu
948: @iftex
949: (See below for VMS.)
950: @end iftex
951:
952: @enumerate
953: @item
1.1.1.4 root 954: Edit @file{Makefile}. If you are using HPUX, or any form of system V,
955: you must make a few changes described in comments at the beginning of
956: the file.
957:
958: @item
959: On a Sequent system, go to the Berkeley universe.
1.1.1.2 root 960:
961: @item
962: Choose configuration files.
963:
964: @itemize @bullet
965: @item
966: Make a symbolic link named @file{config.h} to the top-level
967: config file for the machine you are using (@pxref{Config}). This
968: file is responsible for defining information about the host
969: machine. It includes @file{tm.h}.
970:
971: The file's name should be @file{config-@var{machine}.h}. On VMS,
972: use @file{config-vms.h} rather than @file{config-vax.h}. On the
973: HP 9000 series 300, use @file{config-hp9k3.h} rather than
1.1.1.4 root 974: @file{config-m68k.h}.
1.1.1.2 root 975:
976: If your system does not support symbolic links, you might want to
977: set up @file{config.h} to contain a @samp{#include} command which
978: refers to the appropriate file.
979:
980: @item
981: Make a symbolic link named @file{tm.h} to the machine-description
982: macro file for your machine (its name should be
983: @file{tm-@var{machine}.h}).
984:
985: For the 68000/68020, do not use @file{tm-m68k.h} directly;
986: instead use one of the files @file{tm-sun3.h}, @file{tm-sun2.h},
987: @file{tm-isi68.h}, @file{tm-news800.h} or @file{tm-3b1.h}. Each
988: of those files includes @file{tm-m68k.h} but sets up a few things
989: differently as appropriate to the specific model of
990: machine.@refill
991:
992: There are two files you can use for a 680x0 running HPUX:
993: @file{tm-hp9k320.h} and @file{tm-hp9k320g.h}. Use the former if
994: you are installing GNU CC alone. The latter is for another option
995: where GNU CC together with the GNU assembler, linker, debugger
996: and other utilities are used to replace all of HPUX that deals
997: with compilation. Not all of the pieces of GNU software needed for
998: this mode of operation are as yet in distribution; full instructions
999: will appear here in the future.@refill
1000:
1.1.1.4 root 1001: For the vax, use @file{tm-vax.h} on Unix, or @file{tm-vms.h} on
1002: VMS.@refill
1003:
1.1.1.5 ! root 1004: For the SPARC, use @file{tm-sparc.h}. Note that the SPARC support
! 1005: @strong{has a fatal bug}; to use it, you will have to debug it.
1.1.1.4 root 1006:
1.1.1.2 root 1007: For the 32000, use @file{tm-sequent.h} if you are using a Sequent
1.1.1.3 root 1008: machine, or @file{tm-encore.h} for an Encore machine; otherwise,
1.1.1.4 root 1009: perhaps @file{tm-ns32k.h} will work for you. If you are trying to use
1010: GNU CC on GENIX, you may need to get the version of @code{malloc} from
1011: GNU Emacs instead of the system library version, and you probably need
1012: to cause the following assembler code to be executed in @file{crt0.o}
1013: in order to run the GNU CC output:
1.1.1.2 root 1014:
1.1.1.4 root 1015: @example
1016: lprd sb,$0
1017: sprd mod,r0
1018: movqd $0,0(r0)
1019: @end example
1020:
1021: Note that Encore systems are supported only under BSD.
1.1.1.2 root 1022:
1023: @item
1024: Make a symbolic link named @file{md} to the machine description
1025: pattern file (its name should be @file{@var{machine}.md}).
1026:
1027: @item
1028: Make a symbolic link named @file{aux-output.c} to the output
1029: subroutine file for your machine (its name should be
1030: @file{output-@var{machine}.c}).
1031: @end itemize
1032:
1033: @item
1034: Make sure the Bison parser generator is installed. (This is
1.1.1.4 root 1035: unnecessary if the Bison output files @file{parse.tab.c} and
1036: @file{cexp.c} are more recent than @file{parse.y} and @file{cexp.y}
1037: and you do not plan to change the @samp{.y} files.)
1.1.1.2 root 1038:
1039: Note that if you have an old version of Bison you may get an error
1040: from the line with the @samp{%expect} directive. If so, simply remove
1041: that line from @file{parse.y} and proceed.
1042:
1043: @item
1044: If you are using a Sun, make sure the environment variable
1045: @code{FLOAT_OPTION} is not set. If this option were set to
1046: @code{f68881} when @file{gnulib} is compiled, the resulting code would
1047: demand to be linked with a special startup file and will not link
1048: properly without special pains.
1049:
1050: @item
1051: Build the compiler. Just type @samp{make} in the compiler directory.
1052:
1053: @item
1054: Move the first-stage object files and executables into a subdirectory
1055: with this command:
1056:
1057: @example
1058: make stage1
1059: @end example
1060:
1061: The files are moved into a subdirectory named @file{stage1}.
1062: Once installation is complete, you may wish to delete these files
1063: with @code{rm -r stage1}.
1064:
1065: @item
1066: Recompile the compiler with itself, with this command:
1067:
1068: @example
1069: make CC=stage1/gcc CFLAGS="-g -O -Bstage1/"
1070: @end example
1071:
1072: On a 68000 or 68020 system lacking floating point hardware,
1073: unless you have selected a @file{tm.h} file that expects by default
1074: that there is no such hardware, do this instead:
1075:
1076: @example
1077: make CC=stage1/gcc CFLAGS="-g -O -Bstage1/ -msoft-float"
1078: @end example
1079:
1080: @item
1081: If you wish to test the compiler by compiling it with itself one more
1082: time, do this:
1083:
1084: @example
1085: make stage2
1086: make CC=stage2/gcc CFLAGS="-g -O -Bstage2/"
1087: foreach file (*.o)
1088: cmp $file stage2/$file
1089: end
1090: @end example
1091:
1092: This will notify you if any of these stage 3 object files differs from
1093: those of stage 2. Any difference, no matter how innocuous, indicates
1094: that the stage 2 compiler has compiled GNU CC incorrectly, and is
1095: therefore a potentially serious bug which you should investigate and
1096: report (@pxref{Bugs}).
1097:
1098: @item
1099: Install the compiler driver, the compiler's passes and run-time support.
1100: You can use the following command:
1101:
1102: @example
1103: make install
1104: @end example
1105:
1106: @noindent
1107: This copies the files @file{cc1}, @file{cpp} and @file{gnulib} to
1108: files @file{gcc-cc1}, @file{gcc-cpp} and @file{gcc-gnulib} in
1109: directory @file{/usr/local/lib}, which is where the compiler driver
1110: program looks for them. It also copies the driver program @file{gcc}
1111: into the directory @file{/usr/local}, so that it appears in typical
1112: execution search paths.@refill
1113:
1114: @strong{Warning: the GNU CPP may not work for @file{ioctl.h},
1115: @file{ttychars.h} and other system header files unless the
1116: @samp{-traditional} option is used.} The bug is in the header files:
1117: at least on some machines, they rely on behavior that is incompatible
1118: with ANSI C. This behavior consists of substituting for macro
1119: argument names when they appear inside of character constants. The
1120: @samp{-traditional} option tells GNU CC to behave the way these
1121: headers expect.
1122:
1123: Because of this problem, you might prefer to configure GNU CC to use
1124: the system's own C preprocessor. To do so, make the file
1125: @file{/usr/local/lib/gcc-cpp} a link to @file{/lib/cpp}.
1126:
1127: Alternatively, on Sun systems and 4.3BSD at least, you can correct the
1128: include files by running the shell script @file{fixincludes}. This
1.1.1.5 ! root 1129: installs modified, corrected copies of the files @file{ioctl.h},
! 1130: @file{ttychars.h} and many others, in a special directory where only
! 1131: GNU CC will normally look for them.
1.1.1.2 root 1132:
1.1.1.5 ! root 1133: See the file @file{fixincludes} for a list of all the files we know to
! 1134: require correction.
1.1.1.2 root 1135: @end enumerate
1136:
1137: If you cannot install the compiler's passes and run-time support in
1138: @file{/usr/local/lib}, you can alternatively use the @samp{-B} option to
1139: specify a prefix by which they may be found. The compiler concatenates
1140: the prefix with the names @file{cpp}, @file{cc1} and @file{gnulib}.
1141: Thus, you can put the files in a directory @file{/usr/foo/gcc} and
1142: specify @samp{-B/usr/foo/gcc/} when you run GNU CC.
1143:
1.1.1.4 root 1144: Also, you can specify an alternative default directory for these files
1145: by setting the Make variable @code{libdir} when you make GNU CC.
1146:
1.1.1.2 root 1147: @node VMS Install,, Installation, Installation
1148: @section Installing GNU CC on VMS
1149:
1.1.1.3 root 1150: The VMS version of GNU CC is distributed in an unusual tape format which
1151: consists of several tape files. The first is a command file; the second is
1152: an executable program which reads Unix tar format; the third is another
1153: command file which uses this program to read the remainder of the tape.
1154:
1155: To load the tape, it suffices to mount it @samp{/foreign} and then do
1156: @samp{@@mta0:} to execute the command file at the beginning of the tape.
1157:
1158: The tape contains executables and object files as well as sources, so no
1159: compilation is necessary unless you change the sources. (This is a good
1160: thing, since you probably don't have any other C compiler.) If you must
1161: recompile, here is how:
1.1.1.2 root 1162:
1163: @enumerate
1164: @item
1165: Copy the file @file{tm-vms.h} to @file{tm.h}, @file{config-vms.h} to
1166: @file{config.h}, @file{vax.md} to @file{md.} and @file{output-vax.c}
1167: to @file{aux-output.c}.@refill
1168:
1169: @item
1170: Type @samp{@@make} to do recompile everything.
1171: @end enumerate
1172:
1.1.1.3 root 1173: To install the @samp{GCC} command so you can use the compiler easily, in
1174: the same manner as you use the VMS C compiler, you must install the VMS CLD
1175: file for GNU CC as follows:
1176:
1177: @enumerate
1178: @item
1179: Define the VMS logical names @samp{GNU_CC} and @samp{GNU_CC_INCLUDE}
1180: to point to the directories where the GNU CC executables
1181: (@samp{gcc-cpp}, @samp{gcc-cc1}, etc.) and the C include files are
1182: kept. This should be done with the commands:@refill
1183:
1184: @example
1185: $ assign /super /system disk:[gcc] gnu_cc
1186: $ assign /super /system disk:[gcc.include] gnu_cc_include
1187: @end example
1188:
1189: @noindent
1190: with the appropriate disk and directory names. These commands can be
1191: placed in your system startup file so they will be executed whenever
1192: the machine is rebooted.
1193:
1194: @item
1195: Install the @samp{GCC} command with the command line:
1196:
1197: @example
1198: $ set command /table=sys$library:dcltables gnu_cc:gcc
1199: @end example
1200:
1201: @noindent
1202: Now you can invoke the compiler with a command like @samp{gcc /verbose
1203: file.c}, which is equivalent to the command @samp{gcc -v -c file.c} in
1204: Unix.
1205: @end enumerate
1206:
1207: @node Trouble, Incompatibilities, Installation, Top
1208: @chapter Trouble in Installation
1209:
1210: Here are some of the things that have caused trouble for people installing
1211: GNU CC.
1212:
1.1.1.4 root 1213: @itemize @bullet
1.1.1.3 root 1214: @item
1215: On certain systems, defining certain environment variables such as
1216: @samp{CC} can interfere with the functioning of @code{make}.
1.1.1.4 root 1217:
1218: @item
1219: Cross compilation can run into trouble for certain machines because
1220: some target machines' assemblers require floating point numbers to be
1221: written as @emph{integer} constants in certain contexts.
1222:
1223: The compiler writes these integer constants by examining the floating
1224: point value as an integer and printing that integer, because this is
1225: simple to write and independent of the details of the floating point
1226: representation. But this does not work if the compiler is running on
1227: a different machine with an incompatible floating point format, or
1228: even a different byte-ordering.
1229:
1230: It is possible to fix this by writing machine-independent code which
1231: understands the floating point representation of the target machine.
1232: I am not interested in doing that much work to compensate for bugs
1233: in assemblers.
1.1.1.3 root 1234: @end itemize
1235:
1236: @node Incompatibilities, Extensions, Trouble, Top
1.1.1.2 root 1237: @chapter Incompatibilities of GNU CC
1238:
1239: There are several noteworthy incompatibilities between GNU C and most
1240: existing (non-ANSI) versions of C.
1241:
1242: Ultimately our intention is that the @samp{-traditional} option will
1243: eliminate most of these incompatibilities by telling GNU C to behave
1244: like the other C compilers.
1245:
1246: @itemize @bullet
1247: @item
1248: GNU CC normally makes string constants read-only. If several
1249: identical-looking string constants are used, GNU CC stores only one
1250: copy of the string.
1251:
1252: One consequence is that you cannot call @code{mktemp} with a string
1253: constant argument. The function @code{mktemp} always alters the
1254: string its argument points to.
1255:
1256: Another consequence is that @code{sscanf} does not work on some
1257: systems when passed a string constant as its format control string.
1258: This is because @code{sscanf} incorrectly tries to write into the
1259: string constant.
1260:
1261: The best solution to these problems is to change the program to use
1262: @code{char}-array variables with initialization strings for these
1263: purposes instead of string constants. But if this is not possible,
1264: you can use the @samp{-fwritable-strings} flag, which directs GNU CC
1265: to handle string constants the same way most C compilers do.
1266:
1267: @item
1268: GNU CC does not substitute macro arguments when they appear inside of
1269: string constants. For example, the following macro in GNU CC
1270:
1271: @example
1272: #define foo(a) "a"
1273: @end example
1274:
1275: @noindent
1276: will produce output @samp{"a"} regardless of what the argument @var{a} is.
1277:
1278: The @samp{-traditional} option directs GNU CC to handle such cases
1279: (among others) in the old-fashioned (non-ANSI) fashion.
1280:
1281: @item
1282: When you use @code{setjmp} and @code{longjmp}, the only automatic
1283: variables guaranteed to remain valid are those declared
1284: @code{volatile}. This is a consequence of automatic register
1285: allocation. Consider this function:
1286:
1287: @example
1288: jmp_buf j;
1289:
1290: foo ()
1291: @{
1292: int a, b;
1293:
1294: a = fun1 ();
1295: if (setjmp (j))
1296: return a;
1297:
1298: a = fun2 ();
1299: /* @r{@code{longjmp (j)} may be occur in @code{fun3}.} */
1300: return a + fun3 ();
1301: @}
1302: @end example
1303:
1304: Here @code{a} may or may not be restored to its first value when the
1305: @code{longjmp} occurs. If @code{a} is allocated in a register, then
1306: its first value is restored; otherwise, it keeps the last value stored
1307: in it.
1308:
1309: If you use the @samp{-W} option with the @samp{-O} option, you will
1310: get a warning when GNU CC thinks such a problem might be possible.
1311:
1312: @item
1313: Declarations of external variables and functions within a block apply
1314: only to the block containing the declaration. In other words, they
1315: have the same scope as any other declaration in the same place.
1316:
1317: In some other C compilers, a @code{extern} declaration affects all the
1318: rest of the file even if it happens within a block.
1319:
1320: The @samp{-traditional} option directs GNU C to treat all @code{extern}
1321: declarations as global, like traditional compilers.
1322:
1323: @item
1324: In traditional C, you can combine @code{long}, etc., with a typedef name,
1325: as shown here:
1326:
1327: @example
1328: typedef int foo;
1329: typedef long foo bar;
1330: @end example
1331:
1332: In ANSI C, this is not allowed: @code{long} and other type modifiers
1333: require an explicit @code{int}. Because this criterion is expressed
1334: by Bison grammar rules rather than C code, the @samp{-traditional}
1335: flag cannot alter it.
1336:
1337: @item
1338: When compiling functions that return structures or unions, GNU CC
1339: output code uses a method different from that used on most versions of
1340: Unix. As a result, code compiled with GNU CC cannot call a
1341: structure-returning function compiled with PCC, and vice versa.
1342:
1343: The method used by GCC is as follows: a structure or union which is 1,
1344: 2, 4 or 8 bytes long is returned like a scalar. A structure or union
1345: with any other size is stored into an address supplied by the caller
1346: in a special, fixed register.
1347:
1348: PCC usually handles all sizes of structures and unions by returning
1349: the address of a block of static storage containing the value. This
1350: method is not used in GCC because it is slower and nonreentrant.
1351:
1352: On systems where PCC works this way, you may be able to make GCC-compiled
1353: code call such functions that were compiled with PCC by declaring them
1354: to return a pointer to the structure or union instead of the structure
1355: or union itself. For example, instead of this:
1356:
1357: @example
1358: struct foo nextfoo ();
1359: @end example
1360:
1361: @noindent
1362: write this:
1363:
1364: @example
1365: struct foo *nextfoo ();
1366: #define nextfoo *nextfoo
1367: @end example
1368:
1369: @noindent
1.1.1.4 root 1370: (Note that this assumes you are using the GNU preprocessor and not
1371: @samp{-traditional}, so that the ANSI antirecursion rules for macro
1372: expansions are effective.)
1.1.1.2 root 1373: @end itemize
1374:
1375: @node Extensions, Bugs, Incompatibilities, Top
1376: @chapter GNU Extensions to the C Language
1377:
1378: GNU C provides several language features not found in ANSI standard C.
1379: (The @samp{-pedantic} option directs GNU CC to print a warning message if
1380: any of these features is used.) To test for the availability of these
1381: features in conditional compilation, check for a predefined macro
1382: @code{__GNUC__}, which is always defined under GNU CC.
1383:
1384: @menu
1385: * Statement Exprs:: Putting statements and declarations inside expressions.
1386: * Naming Types:: Giving a name to the type of some expression.
1387: * Typeof:: @code{typeof}: referring to the type of an expression.
1388: * Lvalues:: Using @samp{?:}, @samp{,} and casts in lvalues.
1389: * Conditionals:: Omitting the middle operand of a @samp{?:} expression.
1390: * Zero-Length:: Zero-length arrays.
1391: * Variable-Length:: Arrays whose length is computed at run time.
1392: * Subscripting:: Any array can be subscripted, even if not an lvalue.
1393: * Pointer Arith:: Arithmetic on @code{void}-pointers and function pointers.
1394: * Constructors:: Constructor expressions give structures, unions
1395: or arrays as values.
1396: * Dollar Signs:: Dollar sign is allowed in identifiers.
1397: * Alignment:: Inquiring about the alignment of a type or variable.
1398: * Inline:: Defining inline functions (as fast as macros).
1399: * Extended Asm:: Assembler instructions with C expressions as operands.
1400: (With them you can define ``built-in'' functions.)
1401: * Asm Labels:: Specifying the assembler name to use for a C symbol.
1402: @end menu
1403:
1404: @node Statement Exprs, Naming Types, Extensions, Extensions
1405: @section Statements and Declarations inside of Expressions
1406:
1407: A compound statement in parentheses may appear inside an expression in GNU
1408: C. This allows you to declare variables within an expression. For
1409: example:
1410:
1411: @example
1412: (@{ int y = foo (); int z;
1413: if (y > 0) z = y;
1414: else z = - y;
1415: z; @})
1416: @end example
1417:
1418: @noindent
1419: is a valid (though slightly more complex than necessary) expression
1420: for the absolute value of @code{foo ()}.
1421:
1422: This feature is especially useful in making macro definitions ``safe'' (so
1423: that they evaluate each operand exactly once). For example, the
1424: ``maximum'' function is commonly defined as a macro in standard C as
1425: follows:
1426:
1427: @example
1428: #define max(a,b) ((a) > (b) ? (a) : (b))
1429: @end example
1430:
1431: @noindent
1432: But this definition computes either @var{a} or @var{b} twice, with bad
1433: results if the operand has side effects. In GNU C, if you know the
1434: type of the operands (here let's assume @code{int}), you can define
1435: the macro safely as follows:
1436:
1437: @example
1438: #define maxint(a,b) \
1439: (@{int _a = (a), _b = (b); _a > _b ? _a : _b; @})
1440: @end example
1441:
1442: Embedded statements are not allowed in constant expressions, such as
1443: the value of an enumeration constant, the width of a bit field, or
1444: the initial value of a static variable.
1445:
1446: If you don't know the type of the operand, you can still do this, but you
1447: must use @code{typeof} (@pxref{Typeof}) or type naming (@pxref{Naming
1448: Types}).
1449:
1450: @node Naming Types, Typeof, Statement Exprs, Extensions
1451: @section Naming an Expression's Type
1452:
1453: You can give a name to the type of an expression using a @code{typedef}
1454: declaration with an initializer. Here is how to define @var{name} as a
1455: type name for the type of @var{exp}:
1456:
1457: @example
1458: typedef @var{name} = @var{exp};
1459: @end example
1460:
1461: This is useful in conjunction with the statements-within-expressions
1462: feature. Here is how the two together can be used to define a safe
1463: ``maximum'' macro that operates on any arithmetic type:
1464:
1465: @example
1466: #define max(a,b) \
1467: (@{typedef _ta = (a), _tb = (b); \
1468: _ta _a = (a); _tb _b = (b); \
1469: _a > _b ? _a : _b; @})
1470: @end example
1471:
1472: The reason for using names that start with underscores for the local
1473: variables is to avoid conflicts with variable names that occur within the
1474: expressions that are substituted for @code{a} and @code{b}. Eventually we
1475: hope to design a new form of declaration syntax that allows you to declare
1476: variables whose scopes start only after their initializers; this will be a
1477: more reliable way to prevent such conflicts.
1478:
1479: @node Typeof, Lvalues, Naming Types, Extensions
1480: @section Referring to a Type with @code{typeof}
1481:
1482: Another way to refer to the type of an expression is with @code{typeof}.
1483: The syntax of using of this keyword looks like @code{sizeof}, but the
1484: construct acts semantically like a type name defined with @code{typedef}.
1485:
1486: There are two ways of writing the argument to @code{typeof}: with an
1487: expression or with a type. Here is an example with an expression:
1488:
1489: @example
1490: typeof (x[0](1))
1491: @end example
1492:
1493: @noindent
1494: This assumes that @code{x} is an array of functions; the type described
1495: is that of the values of the functions.
1496:
1497: Here is an example with a typename as the argument:
1498:
1499: @example
1500: typeof (int *)
1501: @end example
1502:
1503: @noindent
1504: Here the type described is that of pointers to @code{int}.
1505:
1506: A @code{typeof}-construct can be used anywhere a typedef name could be
1507: used. For example, you can use it in a declaration, in a cast, or inside
1508: of @code{sizeof} or @code{typeof}.
1509:
1510: @itemize @bullet
1511: @item
1512: This declares @code{y} with the type of what @code{x} points to.
1513:
1514: @example
1515: typeof (*x) y;
1516: @end example
1517:
1518: @item
1519: This declares @code{y} as an array of such values.
1520:
1521: @example
1522: typeof (*x) y[4];
1523: @end example
1524:
1525: @item
1526: This declares @code{y} as an array of pointers to characters:
1527:
1528: @example
1529: typeof (typeof (char *)[4]) y;
1530: @end example
1531:
1532: @noindent
1533: It is equivalent to the following traditional C declaration:
1534:
1535: @example
1536: char *y[4];
1537: @end example
1538:
1539: To see the meaning of the declaration using @code{typeof}, and why it
1540: might be a useful way to write, let's rewrite it with these macros:
1541:
1542: @example
1543: #define pointer(T) typeof(T *)
1544: #define array(T, N) typeof(T [N])
1545: @end example
1546:
1547: @noindent
1548: Now the declaration can be rewritten this way:
1549:
1550: @example
1551: array (pointer (char), 4) y;
1552: @end example
1553:
1554: @noindent
1555: Thus, @samp{array (pointer (char), 4)} is the type of arrays of 4
1556: pointers to @code{char}.
1557: @end itemize
1558:
1559: @node Lvalues, Conditionals, Typeof, Extensions
1560: @section Generalized Lvalues
1561:
1562: Compound expressions, conditional expressions and casts are allowed as
1563: lvalues provided their operands are lvalues. This means that you can take
1564: their addresses or store values into them.
1565:
1566: For example, a compound expression can be assigned, provided the last
1567: expression in the sequence is an lvalue. These two expressions are
1568: equivalent:
1569:
1570: @example
1571: (a, b) += 5
1572: a, (b += 5)
1573: @end example
1574:
1575: Similarly, the address of the compound expression can be taken. These two
1576: expressions are equivalent:
1577:
1578: @example
1579: &(a, b)
1580: a, &b
1581: @end example
1582:
1583: A conditional expression is a valid lvalue if its type is not void and the
1584: true and false branches are both valid lvalues. For example, these two
1585: expressions are equivalent:
1586:
1587: @example
1588: (a ? b : c) = 5
1589: (a ? b = 5 : (c = 5))
1590: @end example
1591:
1592: A cast is a valid lvalue if its operand is valid. Taking the address of
1593: the cast is the same as taking the address without a cast, except for the
1594: type of the result. For example, these two expressions are equivalent (but
1595: the second may be valid when the type of @samp{a} does not permit a cast to
1596: @samp{int *}).
1597:
1598: @example
1599: &(int *)a
1600: (int **)&a
1601: @end example
1602:
1603: A simple assignment whose left-hand side is a cast works by converting the
1604: right-hand side first to the specified type, then to the type of the inner
1605: left-hand side expression. After this is stored, the value is converter
1606: back to the specified type to become the value of the assignment. Thus, if
1607: @samp{a} has type @samp{char *}, the following two expressions are
1608: equivalent:
1609:
1610: @example
1611: (int)a = 5
1612: (int)(a = (char *)5)
1613: @end example
1614:
1615: An assignment-with-arithmetic operation such as @samp{+=} applied to a cast
1616: performs the arithmetic using the type resulting from the cast, and then
1617: continues as in the previous case. Therefore, these two expressions are
1618: equivalent:
1619:
1620: @example
1621: (int)a += 5
1622: (int)(a = (char *) ((int)a + 5))
1623: @end example
1624:
1625: @node Conditionals, Zero-Length, Lvalues, Extensions
1626: @section Conditional Expressions with Omitted Middle-Operands
1627:
1628: The middle operand in a conditional expression may be omitted. Then
1629: if the first operand is nonzero, its value is the value of the conditional
1630: expression.
1631:
1632: Therefore, the expression
1633:
1634: @example
1635: x ? : y
1636: @end example
1637:
1638: @noindent
1639: has the value of @code{x} if that is nonzero; otherwise, the value of
1640: @code{y}.
1641:
1642: This example is perfectly equivalent to
1643:
1644: @example
1645: x ? x : y
1646: @end example
1647:
1648: @noindent
1649: In this simple case, the ability to omit the middle operand is not
1650: especially useful. When it becomes useful is when the first operand does,
1651: or may (if it is a macro argument), contain a side effect. Then repeating
1652: the operand in the middle would perform the side effect twice. Omitting
1653: the middle operand uses the value already computed without the undesirable
1654: effects of recomputing it.
1655:
1656: @node Zero-Length, Variable-Length, Conditionals, Extensions
1657: @section Arrays of Length Zero
1658:
1659: Zero-length arrays are allowed in GNU C. They are very useful as the last
1660: element of a structure which is really a header for a variable-length
1661: object:
1662:
1663: @example
1664: struct line @{
1665: int length;
1666: char contents[0];
1667: @};
1668:
1669: @{
1670: struct line *thisline
1671: = (struct line *) malloc (sizeof (struct line) + this_length);
1672: thisline->length = thislength;
1673: @}
1674: @end example
1675:
1676: In standard C, you would have to give @code{contents} a length of 1, which
1677: means either you waste space or complicate the argument to @code{malloc}.
1678:
1679: @node Variable-Length, Subscripting, Zero-Length, Extensions
1680: @section Arrays of Variable Length
1681:
1682: Variable-length automatic arrays are allowed in GNU C. These arrays are
1683: declared like any other automatic arrays, but with a length that is not a
1684: constant expression. The storage is allocated at that time and
1685: deallocated when the brace-level is exited. For example:
1686:
1687: @example
1688: FILE *concat_fopen (char *s1, char *s2, char *mode)
1689: @{
1690: char str[strlen (s1) + strlen (s2) + 1];
1691: strcpy (str, s1);
1692: strcat (str, s2);
1693: return fopen (str, mode);
1694: @}
1695: @end example
1696:
1697: You can also define structure types containing variable-length arrays, and
1698: use them even for arguments or function values, as shown here:
1699:
1700: @example
1701: int foo;
1702:
1703: struct entry
1704: @{
1705: char data[foo];
1706: @};
1707:
1708: struct entry
1709: tester (struct entry arg)
1710: @{
1711: struct entry new;
1712: int i;
1713: for (i = 0; i < foo; i++)
1714: new.data[i] = arg.data[i] + 1;
1715: return new;
1716: @}
1717: @end example
1718:
1719: @noindent
1720: (Eventually there will be a way to say that the size of the array is
1721: another member of the same structure.)
1722:
1723: The length of an array is computed on entry to the brace-level where the
1724: array is declared and is remembered for the scope of the array in case you
1725: access it with @code{sizeof}.
1726:
1727: Jumping or breaking out of the scope of the array name will also deallocate
1728: the storage. Jumping into the scope is not allowed; you will get an error
1729: message for it.
1730:
1731: You can use the function @code{alloca} to get an effect much like
1732: variable-length arrays. The function @code{alloca} is available in
1733: many other C implementations (but not in all). On the other hand,
1734: variable-length arrays are more elegant.
1735:
1736: There are other differences between these two methods. Space allocated
1737: with @code{alloca} exists until the containing @emph{function} returns.
1738: The space for a variable-length array is deallocated as soon as the array
1739: name's scope ends. (If you use both variable-length arrays and
1740: @code{alloca} in the same function, deallocation of a variable-length array
1741: will also deallocate anything more recently allocated with @code{alloca}.)
1742:
1743: @node Subscripting, Pointer Arith, Variable-Length, Extensions
1744: @section Non-Lvalue Arrays May Have Subscripts
1745:
1746: Subscripting is allowed on arrays that are not lvalues, even though the
1747: unary @samp{&} operator is not. For example, this is valid in GNU C though
1748: not valid in other C dialects:
1749:
1750: @example
1751: struct foo @{int a[4];@};
1752:
1753: struct foo f();
1754:
1755: bar (int index)
1756: @{
1757: return f().a[index];
1758: @}
1759: @end example
1760:
1761: @node Pointer Arith, Initializers, Subscripting, Extensions
1762: @section Arithmetic on @code{void}-Pointers and Function Pointers
1763:
1764: In GNU C, addition and subtraction operations are supported on pointers to
1765: @code{void} and on pointers to functions. This is done by treating the
1766: size of a @code{void} or of a function as 1.
1767:
1768: A consequence of this is that @code{sizeof} is also allowed on @code{void}
1769: and on function types, and returns 1.
1770:
1771: @node Initializers, Constructors, Pointer Arith, Extensions
1772: @section Non-Constant Initializers
1773:
1774: The elements of an aggregate initializer are not required to be constant
1775: expressions in GNU C. Here is an example of an initializer with run-time
1776: varying elements:
1777:
1778: @example
1779: foo (float f, float g)
1780: @{
1781: float beat_freqs[2] = @{ f-g, f+g @};
1782: @dots{}
1783: @}
1784: @end example
1785:
1786: @node Constructors, Dollar Signs, Initializers, Extensions
1787: @section Constructor Expressions
1788:
1789: GNU C supports constructor expressions. A constructor looks like a cast
1790: containing an initializer. Its value is an object of the type specified in
1791: the cast, containing the elements specified in the initializer. The type
1792: must be a structure, union or array type.
1793:
1794: Assume that @code{struct foo} and @code{structure} are declared as shown:
1795:
1796: @example
1797: struct foo @{int a; char b[2];@} structure;
1798: @end example
1799:
1800: @noindent
1801: Here is an example of constructing a @samp{struct foo} with a constructor:
1802:
1803: @example
1804: structure = ((struct foo) @{x + y, 'a', 0@});
1805: @end example
1806:
1807: @noindent
1808: This is equivalent to writing the following:
1809:
1810: @example
1811: @{
1812: struct foo temp = @{x + y, 'a', 0@};
1813: structure = temp;
1814: @}
1815: @end example
1816:
1817: You can also construct an array. If all the elements of the constructor
1818: are (made up of) simple constant expressions, suitable for use in
1819: initializers, then the constructor is an lvalue and can be coerced to a
1820: pointer to its first element, as shown here:
1821:
1822: @example
1823: char **foo = (char *[]) @{ "x", "y", "z" @};
1824: @end example
1825:
1826: Array constructors whose elements are not simple constants are not very
1827: useful, because the constructor is not an lvalue. There are only two valid
1828: ways to use it: to subscript it, or initialize an array variable with it.
1829: The former is probably slower than a @code{switch} statement, while the
1830: latter does the same thing an ordinary C initializer would do.
1831:
1832: @example
1833: output = ((int[]) @{ 2, x, 28 @}) [input];
1834: @end example
1835:
1836: @node Dollar Signs, Alignment, Constructors, Extensions
1837: @section Dollar Signs in Identifier Names
1838:
1839: In GNU C, you may use dollar signs in identifier names. This is because
1840: many traditional C implementations allow such identifiers.
1841:
1842: @node Alignment, Inline, Dollar Signs, Extensions
1843: @section Inquiring about the Alignment of a Type or Variable
1844:
1845: The keyword @code{__alignof} allows you to inquire about how an object
1846: is aligned, or the minimum alignment usually required by a type. Its
1847: syntax is just like @code{sizeof}.
1848:
1849: For example, if the target machine requires a @code{double} value to be
1850: aligned on an 8-byte boundary, then @code{__alignof (double)} is 8. This
1851: is true on many RISC machines. On more traditional machine designs,
1852: @code{__alignof (double)} is 4 or even 2.
1853:
1854: Some machines never actually require alignment; they allow reference to any
1855: data type even at an odd addresses. For these machines, @code{__alignof}
1856: reports the @emph{recommended} alignment of a type.
1857:
1858: When the operand of @code{__alignof} is an lvalue rather than a type, the
1859: value is the largest alignment that the lvalue is known to have. It may
1860: have this alignment as a result of its data type, or because it is part of
1861: a structure and inherits alignment from that structure. For example, after
1862: this declaration:
1863:
1864: @example
1865: struct foo @{ int x; char y; @} foo1;
1866: @end example
1867:
1868: @noindent
1869: the value of @code{__alignof (foo1.y)} is probably 2 or 4, the same as
1870: @code{__alignof (int)}, even though the data type of @code{foo1.y} does not
1871: itself demand any alignment.@refill
1872:
1873: @node Inline, Extended Asm, Alignment, Extensions
1874: @section An Inline Function is As Fast As a Macro
1875:
1876: By declaring a function @code{inline}, you can direct GNU CC to integrate
1877: that function's code into the code for its callers. This makes execution
1878: faster by eliminating the function-call overhead; in addition, if any of
1879: the actual argument values are constant, their known values may permit
1880: simplifications at compile time so that not all of the inline function's
1881: code needs to be included.
1882:
1883: To declare a function inline, use the @code{inline} keyword in its
1884: declaration, like this:
1885:
1886: @example
1887: inline int
1888: inc (int *a)
1889: @{
1890: (*a)++;
1891: @}
1892: @end example
1893:
1894: You can also make all ``simple enough'' functions inline with the
1895: option @samp{-finline-functions}. Note that certain usages in a
1896: function definition can make it unsuitable for inline substitution.
1897:
1898: When a function is both inline and @code{static}, if all calls to the
1899: function are integrated into the caller, then the function's own assembler
1900: code is never referenced. In this case, GNU CC does not actually output
1901: assembler code for the function, unless you specify the option
1902: @samp{-fkeep-inline-functions}. Some calls cannot be integrated for
1903: various reasons (in particular, calls that precede the function's
1904: definition cannot be integrated, and neither can recursive calls within the
1905: definition). If there is a nonintegrated call, then the function is
1906: compiled to assembler code as usual.
1907:
1908: When an inline function is not @code{static}, then the compiler must assume
1909: that there may be calls from other source files; since a global symbol can
1910: be defined only once in any program, the function must not be defined in
1911: the other source files, so the calls therein cannot be integrated.
1912: Therefore, a non-@code{static} inline function is always compiled on its
1913: own in the usual fashion.
1914:
1915: @node Extended Asm, Asm Labels, Inline, Extensions
1916: @section Assembler Instructions with C Expression Operands
1917:
1918: In an assembler instruction using @code{asm}, you can now specify the
1919: operands of the instruction using C expressions. This means no more
1920: guessing which registers or memory locations will contain the data you want
1921: to use.
1922:
1923: You must specify an assembler instruction template much like what appears
1924: in a machine description, plus an operand constraint string for each
1925: operand.
1926:
1927: For example, here is how to use the 68881's @code{fsinx} instruction:
1928:
1929: @example
1930: asm ("fsinx %1,%0" : "=f" (result) : "f" (angle));
1931: @end example
1932:
1933: @noindent
1934: Here @code{angle} is the C expression for the input operand while
1935: @code{result} is that of the output operand. Each has @samp{"f"} as its
1936: operand constraint, saying that a floating-point register is required. The
1937: constraints use the same language used in the machine description
1938: (@pxref{Constraints}).
1939:
1940: Each operand is described by an operand-constraint string followed by the C
1941: expression in parentheses. A colon separates the assembler template from
1942: the first output operand, and another separates the last output operand
1943: from the first input, if any. Commas separate output operands and separate
1944: inputs. The number of operands is limited to the maximum number of
1945: operands in any instruction pattern in the machine description.
1946:
1.1.1.5 ! root 1947: Output operand expressions must be lvalues; the compiler can check this.
! 1948: The input operands need not be lvalues. The compiler cannot check whether
! 1949: the operands have data types that are reasonable for the instruction being
1.1.1.2 root 1950: executed.
1951:
1.1.1.5 ! root 1952: If there are no output operands, and there are input operands, then you
! 1953: should write two colons in a row where the output operands would go.
! 1954:
1.1.1.2 root 1955: The output operands must be write-only; GNU CC will assume that the values
1956: in these operands before the instruction are dead and need not be
1.1.1.5 ! root 1957: generated. For an operand that is read-write, or in which not all bits are
! 1958: written and the other bits contain useful information, you must logically
! 1959: split its function into two separate operands, one input operand and one
! 1960: write-only output operand. The connection between them is expressed by
! 1961: constraints which say they need to be in the same location when the
! 1962: instruction executes. You can use the same C expression for both operands,
! 1963: or different expressions. For example, here we write the (fictitious)
1.1.1.2 root 1964: @samp{combine} instruction with @code{bar} as its read-only source operand
1965: and @code{foo} as its read-write destination:
1966:
1967: @example
1968: asm ("combine %2,%0" : "=r" (foo) : "0" (foo), "g" (bar));
1969: @end example
1970:
1971: @noindent
1972: The constraint @samp{"0"} for operand 1 says that it must occupy the same
1.1.1.5 ! root 1973: location as operand 0.
! 1974:
! 1975: Unless an output operand has the @samp{&} constraint modifier, GNU CC may
! 1976: allocate it in the same register as an unrelated input operand, on the
! 1977: assumption that the inputs are consumed before the outputs are produced.
! 1978: This assumption may be false if the assembler code actually consists of
! 1979: more than one instruction. In such a case, use @samp{&} for each output
! 1980: operand that may not overlap an input. @xref{Modifiers}.
1.1.1.2 root 1981:
1982: Usually the most convenient way to use these @code{asm} instructions is to
1983: encapsulate them in macros that look like functions. For example,
1984:
1985: @example
1986: #define sin(x) \
1987: (@{ double __value, __arg = (x); \
1988: asm ("fsinx %1,%0": "=f" (__value): "f" (__arg)); \
1989: __value; @})
1990: @end example
1991:
1992: @noindent
1993: Here the variable @code{__arg} is used to make sure that the instruction
1994: operates on a proper @code{double} value, and to accept only those
1995: arguments @code{x} which can convert automatically to a @code{double}.
1996:
1997: Another way to make sure the instruction operates on the correct data type
1998: is to use a cast in the @code{asm}. This is different from using a
1999: variable @code{__arg} in that it converts more different types. For
2000: example, if the desired type were @code{int}, casting the argument to
2001: @code{int} would accept a pointer with no complaint, while assigning the
2002: argument to an @code{int} variable named @code{__arg} would warn about
2003: using a pointer unless the caller explicitly casts it.
2004:
2005: GNU CC assumes for optimization purposes that these instructions have no
2006: side effects except to change the output operands. This does not mean that
2007: instructions with a side effect cannot be used, but you must be careful,
2008: because the compiler may eliminate them if the output operands aren't used,
2009: or move them out of loops, or replace two with one if they constitute a
2010: common subexpression. Also, if your instruction does have a side effect on
2011: a variable that otherwise appears not to change, the old value of the
2012: variable may be reused later if it happens to be found in a register.
2013:
2014: You can prevent an @code{asm} instruction from being deleted, moved or
2015: combined by writing the keyword @code{volatile} after the @code{asm}. For
2016: example:
1.1 root 2017:
1.1.1.2 root 2018: @example
2019: #define set_priority(x) \
2020: asm volatile ("set_priority %1": \
2021: "=m" (*(char *)0): "g" (x))
2022: @end example
1.1 root 2023:
1.1.1.2 root 2024: @noindent
2025: Note that we have supplied an output operand which is not actually used in
2026: the instruction. This is because @code{asm} requires at least one output
2027: operand. This requirement exists for internal implementation reasons and
2028: we might be able to relax it in the future.
2029:
2030: In this case output operand has the additional benefit effect of giving the
2031: appearance of writing in memory. As a result, GNU CC will assume that data
2032: previously fetched from memory must be fetched again if needed again later.
2033: This may be desirable if you have not employed the @code{volatile} keyword
2034: on all the variable declarations that ought to have it.
2035:
2036: @node Asm Labels,,Extended Asm, Extensions
2037: @section Controlling Names Used in Assembler Code
2038:
2039: You can specify the name to be used in the assembler code for a C function
2040: or variable by writing the @code{asm} keyword after the declarator as
2041: follows:
1.1 root 2042:
1.1.1.2 root 2043: @example
2044: int foo asm ("myfoo") = 2;
2045: @end example
1.1 root 2046:
1.1.1.2 root 2047: @noindent
2048: This specifies that the name to be used for the variable @code{foo} in
2049: the assembler code should be @samp{myfoo} rather than the usual
2050: @samp{_foo}.
2051:
2052: On systems where an underscore is normally prepended to the name of a C
2053: function or variable, this feature allows you to define names for the
2054: linker that do not start with an underscore.
2055:
2056: You cannot use @code{asm} in this way in a function @emph{definition}; but
2057: you can get the same effect by writing a declaration for the function
2058: before its definition and putting @code{asm} there, like this:
1.1 root 2059:
1.1.1.2 root 2060: @example
2061: extern func () asm ("FUNC");
1.1 root 2062:
1.1.1.2 root 2063: func (x, y)
2064: int x, y;
2065: @dots{}
2066: @end example
1.1 root 2067:
1.1.1.2 root 2068: It is up to you to make sure that the assembler names you choose do not
2069: conflict with any other assembler symbols. Also, you must not use a
2070: register name; that would produce completely invalid assembler code. GNU
2071: CC does not as yet have the ability to store static variables in registers.
2072: Perhaps that will be added.
2073:
2074: @node Bugs, Portability, Extensions, Top
2075: @chapter Reporting Bugs
2076:
2077: Your bug reports play an essential role in making GNU CC reliable.
2078:
2079: Reporting a bug may help you by bringing a solution to your problem, or it
2080: may not. But in any case the important function of a bug report is to help
2081: the entire community by making the next version of GNU CC work better. Bug
2082: reports are your contribution to the maintenance of GNU CC.
1.1 root 2083:
1.1.1.2 root 2084: In order for a bug report to serve its purpose, you must include the
2085: information that makes for fixing the bug.
1.1 root 2086:
1.1.1.2 root 2087: @menu
2088: * Criteria: Bug Criteria. Have you really found a bug?
2089: * Reporting: Bug Reporting. How to report a bug effectively.
2090: @end menu
1.1 root 2091:
1.1.1.2 root 2092: @node Bug Criteria, Bug Reporting, Bugs, Bugs
2093: @section Have You Found a Bug?
2094:
2095: If you are not sure whether you have found a bug, here are some guidelines:
1.1 root 2096:
2097: @itemize @bullet
2098: @item
1.1.1.2 root 2099: If the compiler gets a fatal signal, for any input whatever, that is a
2100: compiler bug. Reliable compilers never crash.
1.1 root 2101:
2102: @item
1.1.1.2 root 2103: If the compiler produces invalid assembly code, for any input whatever
2104: (except an @code{asm} statement), that is a compiler bug, unless the
2105: compiler reports errors (not just warnings) which would ordinarily
2106: prevent the assembler from being run.
1.1 root 2107:
2108: @item
1.1.1.2 root 2109: If the compiler produces valid assembly code that does not correctly
2110: execute the input source code, that is a compiler bug.
1.1 root 2111:
1.1.1.2 root 2112: However, you must double-check to make sure, because you may have run
2113: into an incompatibility between GNU C and traditional C
2114: (@pxref{Incompatibilities}). These incompatibilities might be considered
1.1.1.3 root 2115: bugs, but they are inescapable consequences of valuable features.
1.1 root 2116:
1.1.1.2 root 2117: Or you may have a program whose behavior is undefined, which happened
2118: by chance to give the desired results with another C compiler.
1.1 root 2119:
1.1.1.2 root 2120: For example, in many nonoptimizing compilers, you can write @samp{x;}
2121: at the end of a function instead of @samp{return x;}, with the same
2122: results. But the value of the function is undefined if @samp{return}
2123: is omitted; it is not a bug when GNU CC produces different results.
1.1 root 2124:
1.1.1.2 root 2125: Problems often result from expressions with two increment operators,
2126: as in @samp{f (*p++, *p++)}. Your previous compiler might have
2127: interpreted that expression the way you intended; GNU CC might
2128: interpret it another way; neither compiler is wrong.
1.1 root 2129:
1.1.1.2 root 2130: After you have localized the error to a single source line, it should
2131: be easy to check for these things. If your program is correct and
2132: well defined, you have found a compiler bug.
1.1 root 2133:
1.1.1.2 root 2134: @item
2135: If the compiler produces an error message for valid input, that is a
2136: compiler bug.
1.1 root 2137:
1.1.1.2 root 2138: Note that the following is not valid input, and the error message for
2139: it is not a bug:
1.1 root 2140:
1.1.1.2 root 2141: @example
2142: int foo (char);
1.1 root 2143:
1.1.1.2 root 2144: int
2145: foo (x)
2146: char x;
2147: @{ @dots{} @}
2148: @end example
1.1 root 2149:
1.1.1.2 root 2150: @noindent
2151: The prototype says to pass a @code{char}, while the definition says to
2152: pass an @code{int} and treat the value as a @code{char}. This is what
2153: the ANSI standard says, and it makes sense.
1.1 root 2154:
1.1.1.2 root 2155: @item
2156: If the compiler does not produce an error message for invalid input,
2157: that is a compiler bug. However, you should note that your idea of
2158: ``invalid input'' might be my idea of ``an extension'' or ``support
2159: for traditional practice''.
1.1 root 2160:
1.1.1.2 root 2161: @item
2162: If you are an experienced user of C compilers, your suggestions
2163: for improvement of GNU CC are welcome in any case.
2164: @end itemize
1.1 root 2165:
1.1.1.2 root 2166: @node Bug Reporting,, Bug Criteria, Bugs
2167: @section How to Report Bugs
1.1 root 2168:
1.1.1.2 root 2169: Send bug reports for GNU C to one of these addresses:
1.1 root 2170:
1.1.1.2 root 2171: @example
2172: bug-gcc@@prep.ai.mit.edu
2173: @{ucbvax|mit-eddie|uunet@}!prep.ai.mit.edu!bug-gcc
2174: @end example
1.1 root 2175:
1.1.1.2 root 2176: As a last resort, snail them to:
1.1 root 2177:
1.1.1.2 root 2178: @example
2179: GNU Compiler Bugs
2180: 545 Tech Sq
2181: Cambridge, MA 02139
2182: @end example
1.1 root 2183:
1.1.1.2 root 2184: The fundamental principle of reporting bugs usefully is this:
2185: @strong{report all the facts}. If you are not sure whether to mention a
2186: fact or leave it out, mention it!
2187:
2188: Often people omit facts because they think they know what causes the
2189: problem and they conclude that some details don't matter. Thus, you might
2190: assume that the name of the variable you use in an example does not matter.
2191: Well, probably it doesn't, but one cannot be sure. Perhaps the bug is a
2192: stray memory reference which happens to fetch from the location where that
2193: name is stored in memory; perhaps, if the name were different, the contents
2194: of that location would fool the compiler into doing the right thing despite
2195: the bug. Play it safe and give an exact example.
1.1 root 2196:
1.1.1.2 root 2197: If you want to enable me to fix the bug, you should include all these
2198: things:
1.1 root 2199:
1.1.1.2 root 2200: @itemize @bullet
2201: @item
2202: The version of GNU CC. You can get this by running it with the
2203: @samp{-v} option.
1.1 root 2204:
1.1.1.2 root 2205: Without this, I won't know whether there is any point in looking for
2206: the bug in the current version of GNU CC.
1.1 root 2207:
1.1.1.2 root 2208: @item
2209: A complete input file that will reproduce the bug. If the bug is in
2210: the C preprocessor, send me a source file and any header files that it
2211: requires. If the bug is in the compiler proper (@file{cc1}), run your
2212: source file through the C preprocessor by doing @samp{gcc -E
2213: @var{sourcefile} > @var{outfile}}, then include the contents of
2214: @var{outfile} in the bug report. (Any @samp{-I}, @samp{-D} or
2215: @samp{-U} options that you used in actual compilation should also be
2216: used when doing this.)
1.1 root 2217:
1.1.1.2 root 2218: A single statement is not enough of an example. In order to compile
2219: it, it must be embedded in a function definition; and the bug might
2220: depend on the details of how this is done.
2221:
2222: Without a real example I can compile, all I can do about your bug
2223: report is wish you luck. It would be futile to try to guess how to
2224: provoke the bug. For example, bugs in register allocation and
2225: reloading frequently depend on every little detail of the function
2226: they happen in.
1.1 root 2227:
2228: @item
1.1.1.2 root 2229: The command arguments you gave GNU CC to compile that example and
2230: observe the bug. For example, did you use @samp{-O}? To guarantee
2231: you won't omit something important, list them all.
2232:
2233: If I were to try to guess the arguments, I would probably guess wrong
2234: and then I would not encounter the bug.
1.1 root 2235:
2236: @item
1.1.1.2 root 2237: The names of the files that you used for @file{tm.h} and @file{md}
2238: when you installed the compiler.
1.1 root 2239:
2240: @item
1.1.1.2 root 2241: The type of machine you are using, and the operating system name and
2242: version number.
1.1 root 2243:
2244: @item
1.1.1.2 root 2245: A description of what behavior you observe that you believe is
2246: incorrect. For example, ``It gets a fatal signal,'' or, ``There is an
2247: incorrect assembler instruction in the output.''
2248:
2249: Of course, if the bug is that the compiler gets a fatal signal, then I
2250: will certainly notice it. But if the bug is incorrect output, I might
2251: not notice unless it is glaringly wrong. I won't study all the
2252: assembler code from a 50-line C program just on the off chance that it
2253: might be wrong.
2254:
2255: Even if the problem you experience is a fatal signal, you should still
2256: say so explicitly. Suppose something strange is going on, such as,
2257: your copy of the compiler is out of synch, or you have encountered a
2258: bug in the C library on your system. (This has happened!) Your copy
2259: might crash and mine would not. If you @i{told} me to expect a crash,
2260: then when mine fails to crash, I would know that the bug was not
2261: happening for me. If you had not told me to expect a crash, then I
2262: would not be able to draw any conclusion from my observations.
2263:
2264: In cases where GNU CC generates incorrect code, if you send me a small
2265: complete sample program I will find the error myself by running the
2266: program under a debugger. If you send me a large example or a part of
2267: a larger program, I cannot do this; you must debug the compiled
2268: program and narrow the problem down to one source line. Tell me which
2269: source line it is, and what you believe is incorrect about the code
2270: generated for that line.
1.1 root 2271:
2272: @item
1.1.1.2 root 2273: If you send me examples of output from GNU CC, please use @samp{-g}
2274: when you make them. The debugging information includes source line
2275: numbers which are essential for correlating the output with the input.
1.1.1.4 root 2276:
2277: @item
2278: If you wish to suggest changes to the GNU CC source, send me context
2279: diffs. If you even discuss something in the GNU CC source, refer to
2280: it by context, not by line number.
2281:
2282: The line numbers in my development sources don't match those in your
2283: sources. They won't tell me anything.
1.1 root 2284: @end itemize
2285:
1.1.1.2 root 2286: Here are some things that are not necessary:
1.1 root 2287:
1.1.1.2 root 2288: @itemize @bullet
1.1 root 2289: @item
1.1.1.2 root 2290: A description of the envelope of the bug.
1.1 root 2291:
1.1.1.2 root 2292: Often people who encounter a bug spend a lot of time investigating
2293: which changes to the input file will make the bug go away and which
2294: changes will not affect it.
1.1 root 2295:
1.1.1.2 root 2296: This is often time consuming and not very useful, because the way I
2297: will find the bug is by running a single example under the debugger
2298: with breakpoints, not by pure deduction from a series of examples.
1.1 root 2299:
1.1.1.2 root 2300: Of course, it can't hurt if you can find a simpler example that
2301: triggers the same bug. Errors in the output will be easier to spot,
2302: running under the debugger will take less time, etc. An easy way
2303: to simplify an example is to delete all the function definitions
2304: except the one where the bug occurs. Those earlier in the file
2305: may be replaced by external declarations.
2306:
2307: However, simplification is not necessary; if you don't want to do
2308: this, report the bug anyway.
1.1 root 2309:
2310: @item
1.1.1.2 root 2311: A patch for the bug.
1.1 root 2312:
1.1.1.2 root 2313: A patch for the bug does help me if it is a good one. But don't omit
2314: the necessary information, such as the test case, because I might see
2315: problems with your patch and decide to fix the problem another way.
1.1 root 2316:
1.1.1.2 root 2317: Sometimes with a program as complicated as GNU CC it is very hard to
2318: construct an example that will make the program go through a certain
2319: point in the code. If you don't send me the example, I won't be able
2320: to verify that the bug is fixed.
1.1 root 2321:
2322: @item
1.1.1.2 root 2323: A guess about what the bug is or what it depends on.
2324:
2325: Such guesses are usually wrong. Even I can't guess right about such
2326: things without using the debugger to find the facts. They also don't
2327: serve a useful purpose.
2328: @end itemize
1.1 root 2329:
1.1.1.2 root 2330: @node Portability, Interface, Bugs, Top
1.1 root 2331: @chapter GNU CC and Portability
2332:
2333: The main goal of GNU CC was to make a good, fast compiler for machines in
2334: the class that the GNU system aims to run on: 32-bit machines that address
2335: 8-bit bytes and have several general registers. Elegance, theoretical
2336: power and simplicity are only secondary.
2337:
2338: GNU CC gets most of the information about the target machine from a machine
2339: description which gives an algebraic formula for each of the machine's
2340: instructions. This is a very clean way to describe the target. But when
2341: the compiler needs information that is difficult to express in this
2342: fashion, I have not hesitated to define an ad-hoc parameter to the machine
2343: description. The purpose of portability is to reduce the total work needed
2344: on the compiler; it was not of interest for its own sake.
2345:
2346: GNU CC does not contain machine dependent code, but it does contain code
2347: that depends on machine parameters such as endianness (whether the most
2348: significant byte has the highest or lowest address of the bytes in a word)
2349: and the availability of autoincrement addressing. In the RTL-generation
2350: pass, it is often necessary to have multiple strategies for generating code
2351: for a particular kind of syntax tree, strategies that are usable for different
2352: combinations of parameters. Often I have not tried to address all possible
2353: cases, but only the common ones or only the ones that I have encountered.
2354: As a result, a new target may require additional strategies. You will know
2355: if this happens because the compiler will call @code{abort}. Fortunately,
1.1.1.2 root 2356: the new strategies can be added in a machine-independent fashion, and will
2357: affect only the target machines that need them.
2358:
2359: @node Interface, Passes, Portability, Top
2360: @chapter Interfacing to GNU CC Output
2361:
2362: GNU CC is normally configured to use the same function calling convention
2363: normally in use on the target system. This is done with the
2364: machine-description macros described (@pxref{Machine Macros}).
2365:
2366: However, returning of structure and union values is done differently.
2367: As a result, functions compiled with PCC returning such types cannot
2368: be called from code compiled with GNU CC, and vice versa. This usually
2369: does not cause trouble because the Unix library routines don't return
2370: structures and unions.
2371:
2372: Structures and unions that are 1, 2, 4 or 8 bytes long are returned in the
2373: same registers used for @code{int} or @code{double} return values. (GNU CC
2374: typically allocates variables of such types in registers also.) Structures
2375: and unions of other sizes are returned by storing them into an address
2376: passed by the caller in a register. This method is faster than the one
2377: normally used by PCC and is also reentrant. The register used for passing
2378: the address is specified by the machine-description macro
2379: @code{STRUCT_VALUE_REGNUM}.
2380:
2381: GNU CC always passes arguments on the stack. At some point it will be
2382: extended to pass arguments in registers, for machines which use that as
2383: the standard calling convention. This will make it possible to use such
2384: a convention on other machines as well. However, that would render it
2385: completely incompatible with PCC. We will probably do this once we
2386: have a complete GNU system so we can compile the libraries with GNU CC.
2387:
2388: If you use @code{longjmp}, beware of automatic variables. ANSI C says that
2389: automatic variables that are not declared @code{volatile} have undefined
2390: values after a @code{longjmp}. And this is all GNU CC promises to do,
2391: because it is very difficult to restore register variables correctly, and
2392: one of GNU CC's features is that it can put variables in registers without
2393: your asking it to.
2394:
2395: If you want a variable to be unaltered by @code{longjmp}, and you don't
2396: want to write @code{volatile} because old C compilers don't accept it,
2397: just take the address of the variable. If a variable's address is ever
2398: taken, even if just to compute it and ignore it, then the variable cannot
2399: go in a register:
2400:
2401: @example
2402: @{
2403: int careful;
2404: &careful;
2405: @dots{}
2406: @}
2407: @end example
1.1 root 2408:
1.1.1.2 root 2409: Code compiled with GNU CC may call certain library routines. The routines
2410: needed on the Vax and 68000 are in the file @file{gnulib.c}. You must
2411: compile this file with the standard C compiler, not with GNU CC, and then
2412: link it with each program you compile with GNU CC. (In actuality, many
2413: programs will not need it.) The usual function call interface is used
2414: for calling the library routines. Some standard parts of the C library,
2415: such as @code{bcopy}, are also called automatically.
2416:
2417: @node Passes, RTL, Interface, Top
1.1 root 2418: @chapter Passes and Files of the Compiler
2419:
2420: The overall control structure of the compiler is in @file{toplev.c}. This
2421: file is responsible for initialization, decoding arguments, opening and
2422: closing files, and sequencing the passes.
2423:
1.1.1.2 root 2424: The parsing pass is invoked only once, to parse the entire input. The RTL
2425: intermediate code for a function is generated as the function is parsed, a
2426: statement at a time. Each statement is read in as a syntax tree and then
2427: converted to RTL; then the storage for the tree for the statement is
2428: reclaimed. Storage for types (and the expressions for their sizes),
2429: declarations, and a representation of the binding contours and how they nest,
2430: remains until the function is finished being compiled; these are all needed
2431: to output the debugging information.
2432:
2433: Each time the parsing pass reads a complete function definition or
2434: top-level declaration, it calls the function
2435: @code{rest_of_compilation} or @code{rest_of_decl_compilation} in
2436: @file{toplev.c}, which are responsible for all further processing
2437: necessary, ending with output of the assembler language. All other
2438: compiler passes run, in sequence, within @code{rest_of_compilation}.
2439: When that function returns from compiling a function definition, the
2440: storage used for that function definition's compilation is entirely
2441: freed, unless it is an inline function (@pxref{Inline}).
1.1 root 2442:
2443: Here is a list of all the passes of the compiler and their source files.
2444: Also included is a description of where debugging dumps can be requested
1.1.1.2 root 2445: with @samp{-d} options.
1.1 root 2446:
2447: @itemize @bullet
2448: @item
2449: Parsing. This pass reads the entire text of a function definition,
1.1.1.2 root 2450: constructing partial syntax trees. This and RTL generation are no longer
2451: truly separate passes (formerly they were), but it is easier to think
2452: of them as separate.
2453:
2454: The tree representation does not entirely follow C syntax, because it is
2455: intended to support other languages as well.
1.1 root 2456:
1.1.1.2 root 2457: C data type analysis is also done in this pass, and every tree node
2458: that represents an expression has a data type attached. Variables are
2459: represented as declaration nodes.
1.1 root 2460:
1.1.1.2 root 2461: Constant folding and associative-law simplifications are also done
2462: during this pass.
1.1 root 2463:
1.1.1.2 root 2464: The source files for parsing are @file{parse.y}, @file{decl.c},
1.1 root 2465: @file{typecheck.c}, @file{stor-layout.c}, @file{fold-const.c}, and
2466: @file{tree.c}. The last three are intended to be language-independent.
2467: There are also header files @file{parse.h}, @file{c-tree.h},
2468: @file{tree.h} and @file{tree.def}. The last two define the format of
1.1.1.2 root 2469: the tree representation.@refill
1.1 root 2470:
2471: @item
1.1.1.2 root 2472: RTL generation. This is the conversion of syntax tree into RTL code.
2473: It is actually done statement-by-statement during parsing, but for
2474: most purposes it can be thought of as a separate pass.
1.1 root 2475:
2476: This is where the bulk of target-parameter-dependent code is found,
2477: since often it is necessary for strategies to apply only when certain
2478: standard kinds of instructions are available. The purpose of named
2479: instruction patterns is to provide this information to the RTL
2480: generation pass.
2481:
2482: Optimization is done in this pass for @code{if}-conditions that are
2483: comparisons, boolean operations or conditional expressions. Tail
2484: recursion is detected at this time also. Decisions are made about how
2485: best to arrange loops and how to output @code{switch} statements.
2486:
1.1.1.2 root 2487: The source files for RTL generation are @file{stmt.c}, @file{expr.c},
1.1 root 2488: @file{explow.c}, @file{expmed.c}, @file{optabs.c} and @file{emit-rtl.c}.
2489: Also, the file @file{insn-emit.c}, generated from the machine description
2490: by the program @code{genemit}, is used in this pass. The header files
1.1.1.2 root 2491: @file{expr.h} is used for communication within this pass.@refill
1.1 root 2492:
1.1.1.2 root 2493: The header files @file{insn-flags.h} and @file{insn-codes.h},
2494: generated from the machine description by the programs @code{genflags}
2495: and @code{gencodes}, tell this pass which standard names are available
2496: for use and which patterns correspond to them.@refill
1.1 root 2497:
2498: Aside from debugging information output, none of the following passes
1.1.1.2 root 2499: refers to the tree structure representation of the function (only
2500: part of which is saved).
1.1 root 2501:
1.1.1.2 root 2502: The decision of whether the function can and should be expanded inline
2503: in its subsequent callers is made at the end of rtl generation. The
2504: function must meet certain criteria, currently related to the size of
2505: the function and the types and number of parameters it has. Note that
2506: this function may contain loops, recursive calls to itself
2507: (tail-recursive functions can be inlined!), gotos, in short, all
2508: constructs supported by GNU CC.
2509:
2510: The option @samp{-dr} causes a debugging dump of the RTL code after
2511: this pass. This dump file's name is made by appending @samp{.rtl} to
2512: the input file name.
1.1 root 2513:
2514: @item
1.1.1.2 root 2515: Jump optimization. This pass simplifies jumps to the following
2516: instruction, jumps across jumps, and jumps to jumps. It deletes
2517: unreferenced labels and unreachable code, except that unreachable code
2518: that contains a loop is not recognized as unreachable in this pass.
2519: (Such loops are deleted later in the basic block analysis.)
1.1 root 2520:
2521: Jump optimization is performed two or three times. The first time is
1.1.1.2 root 2522: immediately following RTL generation. The second time is after CSE,
2523: but only if CSE says repeated jump optimization is needed. The
2524: last time is right before the final pass. That time, cross-jumping
2525: and deletion of no-op move instructions are done together with the
2526: optimizations described above.
1.1 root 2527:
2528: The source file of this pass is @file{jump.c}.
2529:
1.1.1.2 root 2530: The option @samp{-dj} causes a debugging dump of the RTL code after
2531: this pass is run for the first time. This dump file's name is made by
2532: appending @samp{.jump} to the input file name.
1.1 root 2533:
2534: @item
2535: Register scan. This pass finds the first and last use of each
2536: register, as a guide for common subexpression elimination. Its source
2537: is in @file{regclass.c}.
2538:
2539: @item
2540: Common subexpression elimination. This pass also does constant
2541: propagation. Its source file is @file{cse.c}. If constant
2542: propagation causes conditional jumps to become unconditional or to
1.1.1.2 root 2543: become no-ops, jump optimization is run again when CSE is finished.
1.1 root 2544:
1.1.1.2 root 2545: The option @samp{-ds} causes a debugging dump of the RTL code after
1.1 root 2546: this pass. This dump file's name is made by appending @samp{.cse} to
2547: the input file name.
2548:
2549: @item
2550: Loop optimization. This pass moves constant expressions out of loops.
2551: Its source file is @file{loop.c}.
2552:
1.1.1.2 root 2553: The option @samp{-dL} causes a debugging dump of the RTL code after
1.1 root 2554: this pass. This dump file's name is made by appending @samp{.loop} to
2555: the input file name.
2556:
2557: @item
2558: Stupid register allocation is performed at this point in a
2559: nonoptimizing compilation. It does a little data flow analysis as
2560: well. When stupid register allocation is in use, the next pass
2561: executed is the reloading pass; the others in between are skipped.
1.1.1.2 root 2562: The source file is @file{stupid.c}.
1.1 root 2563:
2564: @item
2565: Data flow analysis (@file{flow.c}). This pass divides the program
2566: into basic blocks (and in the process deletes unreachable loops); then
2567: it computes which pseudo-registers are live at each point in the
2568: program, and makes the first instruction that uses a value point at
2569: the instruction that computed the value.
2570:
2571: This pass also deletes computations whose results are never used, and
2572: combines memory references with add or subtract instructions to make
2573: autoincrement or autodecrement addressing.
2574:
1.1.1.2 root 2575: The option @samp{-df} causes a debugging dump of the RTL code after
1.1 root 2576: this pass. This dump file's name is made by appending @samp{.flow} to
2577: the input file name. If stupid register allocation is in use, this
2578: dump file reflects the full results of such allocation.
2579:
2580: @item
2581: Instruction combination (@file{combine.c}). This pass attempts to
2582: combine groups of two or three instructions that are related by data
2583: flow into single instructions. It combines the RTL expressions for
2584: the instructions by substitution, simplifies the result using algebra,
2585: and then attempts to match the result against the machine description.
2586:
1.1.1.2 root 2587: The option @samp{-dc} causes a debugging dump of the RTL code after
1.1 root 2588: this pass. This dump file's name is made by appending @samp{.combine}
2589: to the input file name.
2590:
2591: @item
2592: Register class preferencing. The RTL code is scanned to find out
1.1.1.2 root 2593: which register class is best for each pseudo register. The source
2594: file is @file{regclass.c}.
1.1 root 2595:
2596: @item
2597: Local register allocation (@file{local-alloc.c}). This pass allocates
2598: hard registers to pseudo registers that are used only within one basic
1.1.1.2 root 2599: block. Because the basic block is linear, it can use fast and
2600: powerful techniques to do a very good job.
1.1 root 2601:
1.1.1.2 root 2602: The option @samp{-dl} causes a debugging dump of the RTL code after
1.1 root 2603: this pass. This dump file's name is made by appending @samp{.lreg} to
2604: the input file name.
2605:
2606: @item
2607: Global register allocation (@file{global-alloc.c}). This pass
2608: allocates hard registers for the remaining pseudo registers (those
2609: whose life spans are not contained in one basic block).
2610:
2611: @item
1.1.1.2 root 2612: Reloading. This pass renumbers pseudo registers with the hardware
2613: registers numbers they were allocated. Pseudo registers that did not
2614: get hard registers are replaced with stack slots. Then it finds
2615: instructions that are invalid because a value has failed to end up in
2616: a register, or has ended up in a register of the wrong kind. It fixes
2617: up these instructions by reloading the problematical values
2618: temporarily into registers. Additional instructions are generated to
2619: do the copying.
1.1 root 2620:
2621: Source files are @file{reload.c} and @file{reload1.c}, plus the header
2622: @file{reload.h} used for communication between them.
2623:
1.1.1.2 root 2624: The option @samp{-dg} causes a debugging dump of the RTL code after
1.1 root 2625: this pass. This dump file's name is made by appending @samp{.greg} to
2626: the input file name.
2627:
2628: @item
1.1.1.2 root 2629: Jump optimization is repeated, this time including cross-jumping
2630: and deletion of no-op move instructions. Machine-specific peephole
2631: optimizations are performed at the same time.
2632:
2633: The option @samp{-dJ} causes a debugging dump of the RTL code after
2634: this pass. This dump file's name is made by appending @samp{.jump2}
2635: to the input file name.
1.1 root 2636:
2637: @item
2638: Final. This pass outputs the assembler code for the function. It is
1.1.1.2 root 2639: also responsible for identifying spurious test and compare
2640: instructions. The function entry and exit sequences are generated
2641: directly as assembler code in this pass; they never exist as RTL.
1.1 root 2642:
2643: The source files are @file{final.c} plus @file{insn-output.c}; the
2644: latter is generated automatically from the machine description by the
2645: tool @file{genoutput}. The header file @file{conditions.h} is used
2646: for communication between these files.
2647:
2648: @item
2649: Debugging information output. This is run after final because it must
2650: output the stack slot offsets for pseudo registers that did not get
2651: hard registers. Source files are @file{dbxout.c} for DBX symbol table
2652: format and @file{symout.c} for GDB's own symbol table format.
2653: @end itemize
2654:
2655: Some additional files are used by all or many passes:
2656:
2657: @itemize @bullet
2658: @item
2659: Every pass uses @file{machmode.def}, which defines the machine modes.
2660:
2661: @item
2662: All the passes that work with RTL use the header files @file{rtl.h}
1.1.1.2 root 2663: and @file{rtl.def}, and subroutines in file @file{rtl.c}. The tools
2664: @code{gen*} also use these files to read and work with the machine
2665: description RTL.
1.1 root 2666:
2667: @item
2668: Several passes refer to the header file @file{insn-config.h} which
2669: contains a few parameters (C macro definitions) generated
2670: automatically from the machine description RTL by the tool
2671: @code{genconfig}.
2672:
2673: @item
2674: Several passes use the instruction recognizer, which consists of
2675: @file{recog.c} and @file{recog.h}, plus the files @file{insn-recog.c}
2676: and @file{insn-extract.c} that are generated automatically from the
1.1.1.2 root 2677: machine description by the tools @file{genrecog} and
2678: @file{genextract}.@refill
1.1 root 2679:
2680: @item
1.1.1.2 root 2681: Several passes use the header files @file{regs.h} which defines the
2682: information recorded about pseudo register usage, and @file{basic-block.h}
1.1 root 2683: which defines the information recorded about basic blocks.
2684:
2685: @item
2686: @file{hard-reg-set.h} defines the type @code{HARD_REG_SET}, a bit-vector
2687: with a bit for each hard register, and some macros to manipulate it.
2688: This type is just @code{int} if the machine has few enough hard registers;
2689: otherwise it is an array of @code{int} and some of the macros expand
2690: into loops.
2691: @end itemize
2692:
2693: @node RTL, Machine Desc, Passes, Top
2694: @chapter RTL Representation
2695:
2696: Most of the work of the compiler is done on an intermediate representation
1.1.1.2 root 2697: called register transfer language. In this language, the instructions to be
1.1 root 2698: output are described, pretty much one by one, in an algebraic form that
2699: describes what the instruction does.
2700:
2701: RTL is inspired by Lisp lists. It has both an internal form, made up of
2702: structures that point at other structures, and a textual form that is used
2703: in the machine description and in printed debugging dumps. The textual
2704: form uses nested parentheses to indicate the pointers in the internal form.
2705:
2706: @menu
2707: * RTL Objects:: Expressions vs vectors vs strings vs integers.
2708: * Accessors:: Macros to access expression operands or vector elts.
1.1.1.2 root 2709: * Flags:: Other flags in an RTL expression.
1.1 root 2710: * Machine Modes:: Describing the size and format of a datum.
2711: * Constants:: Expressions with constant values.
2712: * Regs and Memory:: Expressions representing register contents or memory.
2713: * Arithmetic:: Expressions representing arithmetic on other expressions.
2714: * Comparisons:: Expressions representing comparison of expressions.
2715: * Bit Fields:: Expressions representing bit-fields in memory or reg.
2716: * Conversions:: Extending, truncating, floating or fixing.
2717: * RTL Declarations:: Declaring volatility, constancy, etc.
2718: * Side Effects:: Expressions for storing in registers, etc.
2719: * Incdec:: Embedded side-effects for autoincrement addressing.
1.1.1.2 root 2720: * Assembler:: Representing @code{asm} with operands.
1.1 root 2721: * Insns:: Expression types for entire insns.
1.1.1.2 root 2722: * Calls:: RTL representation of function call insns.
1.1 root 2723: * Sharing:: Some expressions are unique; others *must* be copied.
2724: @end menu
2725:
2726: @node RTL Objects, Accessors, RTL, RTL
2727: @section RTL Object Types
2728:
2729: RTL uses four kinds of objects: expressions, integers, strings and vectors.
1.1.1.2 root 2730: Expressions are the most important ones. An RTL expression (``RTX'', for
2731: short) is a C structure, but it is usually referred to with a pointer; a
2732: type that is given the typedef name @code{rtx}.
1.1 root 2733:
2734: An integer is simply an @code{int}, and a string is a @code{char *}.
1.1.1.2 root 2735: Within RTL code, strings appear only inside @samp{symbol_ref} expressions,
2736: but they appear in other contexts in the RTL expressions that make up
1.1 root 2737: machine descriptions. Their written form uses decimal digits.
2738:
2739: A string is a sequence of characters. In core it is represented as a
1.1.1.2 root 2740: @code{char *} in usual C fashion, and it is written in C syntax as well.
1.1 root 2741: However, strings in RTL may never be null. If you write an empty string in
2742: a machine description, it is represented in core as a null pointer rather
2743: than as a pointer to a null character. In certain contexts, these null
2744: pointers instead of strings are valid.
2745:
2746: A vector contains an arbitrary, specified number of pointers to
2747: expressions. The number of elements in the vector is explicitly present in
2748: the vector. The written form of a vector consists of square brackets
2749: (@samp{[@dots{}]}) surrounding the elements, in sequence and with
2750: whitespace separating them. Vectors of length zero are not created; null
2751: pointers are used instead.
2752:
1.1.1.2 root 2753: Expressions are classified by @dfn{expression codes} (also called RTX
2754: codes). The expression code is a name defined in @file{rtl.def}, which is
2755: also (in upper case) a C enumeration constant. The possible expression
2756: codes and their meanings are machine-independent. The code of an RTX can
2757: be extracted with the macro @code{GET_CODE (@var{x})} and altered with
2758: @code{PUT_CODE (@var{x}, @var{newcode})}.
1.1 root 2759:
2760: The expression code determines how many operands the expression contains,
2761: and what kinds of objects they are. In RTL, unlike Lisp, you cannot tell
2762: by looking at an operand what kind of object it is. Instead, you must know
2763: from its context---from the expression code of the containing expression.
1.1.1.2 root 2764: For example, in an expression of code @samp{subreg}, the first operand is
1.1 root 2765: to be regarded as an expression and the second operand as an integer. In
1.1.1.2 root 2766: an expression of code @samp{plus}, there are two operands, both of which
2767: are to be regarded as expressions. In a @samp{symbol_ref} expression,
1.1 root 2768: there is one operand, which is to be regarded as a string.
2769:
2770: Expressions are written as parentheses containing the name of the
2771: expression type, its flags and machine mode if any, and then the operands
2772: of the expression (separated by spaces).
2773:
1.1.1.2 root 2774: Expression code names in the @samp{md} file are written in lower case,
2775: but when they appear in C code they are written in upper case. In this
2776: manual, they are shown as follows: @samp{const_int}.
2777:
1.1 root 2778: In a few contexts a null pointer is valid where an expression is normally
2779: wanted. The written form of this is @samp{(nil)}.
2780:
1.1.1.2 root 2781: @node Accessors, Flags, RTL Objects, RTL
1.1 root 2782: @section Access to Operands
2783:
2784: For each expression type @file{rtl.def} specifies the number of contained
2785: objects and their kinds, with four possibilities: @samp{e} for expression
2786: (actually a pointer to an expression), @samp{i} for integer, @samp{s} for
2787: string, and @samp{E} for vector of expressions. The sequence of letters
2788: for an expression code is called its @dfn{format}. Thus, the format of
1.1.1.2 root 2789: @samp{subreg} is @samp{ei}.@refill
1.1 root 2790:
2791: Two other format characters are used occasionally: @samp{u} and @samp{0}.
2792: @samp{u} is equivalent to @samp{e} except that it is printed differently in
2793: debugging dumps, and @samp{0} means a slot whose contents do not fit any
2794: normal category. @samp{0} slots are not printed at all in dumps, and are
1.1.1.2 root 2795: often used in special ways by small parts of the compiler.@refill
1.1 root 2796:
2797: There are macros to get the number of operands and the format of an
2798: expression code:
2799:
2800: @table @code
2801: @item GET_RTX_LENGTH (@var{code})
1.1.1.2 root 2802: Number of operands of an RTX of code @var{code}.
1.1 root 2803:
2804: @item GET_RTX_FORMAT (@var{code})
1.1.1.2 root 2805: The format of an RTX of code @var{code}, as a C string.
1.1 root 2806: @end table
2807:
2808: Operands of expressions are accessed using the macros @code{XEXP},
2809: @code{XINT} and @code{XSTR}. Each of these macros takes two arguments: an
1.1.1.2 root 2810: expression-pointer (RTX) and an operand number (counting from zero).
2811: Thus,@refill
1.1 root 2812:
2813: @example
1.1.1.2 root 2814: XEXP (@var{x}, 2)
1.1 root 2815: @end example
2816:
2817: @noindent
2818: accesses operand 2 of expression @var{x}, as an expression.
2819:
2820: @example
1.1.1.2 root 2821: XINT (@var{x}, 2)
1.1 root 2822: @end example
2823:
2824: @noindent
2825: accesses the same operand as an integer. @code{XSTR}, used in the same
2826: fashion, would access it as a string.
2827:
2828: Any operand can be accessed as an integer, as an expression or as a string.
2829: You must choose the correct method of access for the kind of value actually
2830: stored in the operand. You would do this based on the expression code of
2831: the containing expression. That is also how you would know how many
2832: operands there are.
2833:
2834: For example, if @var{x} is a @samp{subreg} expression, you know that it has
1.1.1.2 root 2835: two operands which can be correctly accessed as @code{XEXP (@var{x}, 0)}
2836: and @code{XINT (@var{x}, 1)}. If you did @code{XINT (@var{x}, 0)}, you
2837: would get the address of the expression operand but cast as an integer;
2838: that might occasionally be useful, but it would be cleaner to write
2839: @code{(int) XEXP (@var{x}, 0)}. @code{XEXP (@var{x}, 1)} would also
2840: compile without error, and would return the second, integer operand cast as
2841: an expression pointer, which would probably result in a crash when
2842: accessed. Nothing stops you from writing @code{XEXP (@var{x}, 28)} either,
2843: but this will access memory past the end of the expression with
2844: unpredictable results.@refill
1.1 root 2845:
2846: Access to operands which are vectors is more complicated. You can use the
2847: macro @code{XVEC} to get the vector-pointer itself, or the macros
2848: @code{XVECEXP} and @code{XVECLEN} to access the elements and length of a
2849: vector.
2850:
2851: @table @code
2852: @item XVEC (@var{exp}, @var{idx})
2853: Access the vector-pointer which is operand number @var{idx} in @var{exp}.
2854:
2855: @item XVECLEN (@var{exp}, @var{idx})
2856: Access the length (number of elements) in the vector which is
2857: in operand number @var{idx} in @var{exp}. This value is an @code{int}.
2858:
1.1.1.2 root 2859: @item XVECEXP (@var{exp}, @var{idx}, @var{eltnum})
1.1 root 2860: Access element number @var{eltnum} in the vector which is
1.1.1.2 root 2861: in operand number @var{idx} in @var{exp}. This value is an RTX.
1.1 root 2862:
2863: It is up to you to make sure that @var{eltnum} is not negative
2864: and is less than @code{XVECLEN (@var{exp}, @var{idx})}.
2865: @end table
2866:
2867: All the macros defined in this section expand into lvalues and therefore
2868: can be used to assign the operands, lengths and vector elements as well as
2869: to access them.
2870:
1.1.1.2 root 2871: @node Flags, Machine Modes, Accessors, RTL
2872: @section Flags in an RTL Expression
2873:
2874: RTL expressions contain several flags (one-bit bit-fields) that are used
2875: in certain types of expression.
2876:
2877: @table @code
2878: @item used
2879: This flag is used only momentarily, at the end of RTL generation for a
2880: function, to count the number of times an expression appears in insns.
2881: Expressions that appear more than once are copied, according to the
2882: rules for shared structure (@pxref{Sharing}).
2883:
2884: @item volatil
2885: This flag is used in @samp{mem} and @samp{reg} expressions and in insns.
2886: In RTL dump files, it is printed as @samp{/v}.
2887:
2888: In a @samp{mem} expression, it is 1 if the memory reference is volatile.
2889: Volatile memory references may not be deleted, reordered or combined.
2890:
2891: In a @samp{reg} expression, it is 1 if the value is a user-level variable.
2892: 0 indicates an internal compiler temporary.
2893:
2894: In an insn, 1 means the insn has been deleted.
2895:
2896: @item in_struct
2897: This flag is used in @samp{mem} expressions. It is 1 if the memory
2898: datum referred to is all or part of a structure or array; 0 if it is (or
2899: might be) a scalar variable. A reference through a C pointer has 0
2900: because the pointer might point to a scalar variable.
2901:
2902: This information allows the compiler to determine something about possible
2903: cases of aliasing.
2904:
2905: In an RTL dump, this flag is represented as @samp{/s}.
2906:
2907: @item unchanging
2908: This flag is used in @samp{reg} and @samp{mem} expressions. 1 means
2909: that the value of the expression never changes (at least within the
2910: current function).
2911:
2912: In an RTL dump, this flag is represented as @samp{/u}.
2913: @end table
2914:
2915: @node Machine Modes, Constants, Flags, RTL
1.1 root 2916: @section Machine Modes
2917:
2918: A machine mode describes a size of data object and the representation used
2919: for it. In the C code, machine modes are represented by an enumeration
1.1.1.2 root 2920: type, @code{enum machine_mode}, defined in @file{machmode.def}. Each RTL
2921: expression has room for a machine mode and so do certain kinds of tree
2922: expressions (declarations and types, to be precise).
1.1 root 2923:
2924: In debugging dumps and machine descriptions, the machine mode of an RTL
2925: expression is written after the expression code with a colon to separate
2926: them. The letters @samp{mode} which appear at the end of each machine mode
2927: name are omitted. For example, @code{(reg:SI 38)} is a @samp{reg}
2928: expression with machine mode @code{SImode}. If the mode is
2929: @code{VOIDmode}, it is not written at all.
2930:
2931: Here is a table of machine modes.
2932:
2933: @table @code
2934: @item QImode
2935: ``Quarter-Integer'' mode represents a single byte treated as an integer.
2936:
2937: @item HImode
2938: ``Half-Integer'' mode represents a two-byte integer.
2939:
2940: @item SImode
2941: ``Single Integer'' mode represents a four-byte integer.
2942:
2943: @item DImode
2944: ``Double Integer'' mode represents an eight-byte integer.
2945:
2946: @item TImode
2947: ``Tetra Integer'' (?) mode represents a sixteen-byte integer.
2948:
2949: @item SFmode
2950: ``Single Floating'' mode represents a single-precision (four byte) floating
2951: point number.
2952:
2953: @item DFmode
2954: ``Double Floating'' mode represents a double-precision (eight byte) floating
2955: point number.
2956:
2957: @item TFmode
2958: ``Tetra Floating'' mode represents a quadruple-precision (sixteen byte)
2959: floating point number.
2960:
2961: @item BLKmode
2962: ``Block'' mode represents values that are aggregates to which none of
1.1.1.2 root 2963: the other modes apply. In RTL, only memory references can have this mode,
1.1 root 2964: and only if they appear in string-move or vector instructions. On machines
2965: which have no such instructions, @code{BLKmode} will not appear in RTL.
2966:
2967: @item VOIDmode
2968: Void mode means the absence of a mode or an unspecified mode.
1.1.1.2 root 2969: For example, RTL expressions of code @samp{const_int} have mode
1.1 root 2970: @code{VOIDmode} because they can be taken to have whatever mode the context
2971: requires. In debugging dumps of RTL, @code{VOIDmode} is expressed by
2972: the absence of any mode.
2973:
2974: @item EPmode
2975: ``Entry Pointer'' mode is intended to be used for function variables in
2976: Pascal and other block structured languages. Such values contain
2977: both a function address and a static chain pointer for access to
2978: automatic variables of outer levels. This mode is only partially
2979: implemented since C does not use it.
2980:
2981: @item CSImode@r{, @dots{}}
2982: ``Complex Single Integer'' mode stands for a complex number represented
2983: as a pair of @code{SImode} integers. Any of the integer and floating modes
2984: may have @samp{C} prefixed to its name to obtain a complex number mode.
2985: For example, there are @code{CQImode}, @code{CSFmode}, and @code{CDFmode}.
2986: Since C does not support complex numbers, these machine modes are only
2987: partially implemented.
2988:
2989: @item BImode
2990: This is the machine mode of a bit-field in a structure. It is used
2991: only in the syntax tree, never in RTL, and in the syntax tree it appears
2992: only in declaration nodes. In C, it appears only in @code{FIELD_DECL}
2993: nodes for structure fields defined with a bit size.
2994: @end table
2995:
2996: The machine description defines @code{Pmode} as a C macro which expands
2997: into the machine mode used for addresses. Normally this is @code{SImode}.
2998:
2999: The only modes which a machine description @i{must} support are
3000: @code{QImode}, @code{SImode}, @code{SFmode} and @code{DFmode}. The
3001: compiler will attempt to use @code{DImode} for two-word structures and
3002: unions, but it would not be hard to program it to avoid this. Likewise,
3003: you can arrange for the C type @code{short int} to avoid using
3004: @code{HImode}. In the long term it would be desirable to make the set of
3005: available machine modes machine-dependent and eliminate all assumptions
3006: about specific machine modes or their uses from the machine-independent
3007: code of the compiler.
3008:
3009: Here are some C macros that relate to machine modes:
3010:
3011: @table @code
3012: @item GET_MODE (@var{x})
1.1.1.2 root 3013: Returns the machine mode of the RTX @var{x}.
1.1 root 3014:
3015: @item PUT_MODE (@var{x}, @var{newmode})
1.1.1.2 root 3016: Alters the machine mode of the RTX @var{x} to be @var{newmode}.
1.1 root 3017:
3018: @item GET_MODE_SIZE (@var{m})
3019: Returns the size in bytes of a datum of mode @var{m}.
3020:
3021: @item GET_MODE_BITSIZE (@var{m})
3022: Returns the size in bits of a datum of mode @var{m}.
3023:
3024: @item GET_MODE_UNIT_SIZE (@var{m})
3025: Returns the size in bits of the subunits of a datum of mode @var{m}.
3026: This is the same as @code{GET_MODE_SIZE} except in the case of
1.1.1.2 root 3027: complex modes and @code{EPmode}. For them, the unit size is the
1.1 root 3028: size of the real or imaginary part, or the size of the function
3029: pointer or the context pointer.
3030: @end table
3031:
3032: @node Constants, Regs and Memory, Machine Modes, RTL
3033: @section Constant Expression Types
3034:
3035: The simplest RTL expressions are those that represent constant values.
3036:
3037: @table @code
3038: @item (const_int @var{i})
3039: This type of expression represents the integer value @var{i}. @var{i}
3040: is customarily accessed with the macro @code{INTVAL} as in
1.1.1.2 root 3041: @code{INTVAL (@var{exp})}, which is equivalent to @code{XINT (@var{exp}, 0)}.
1.1 root 3042:
3043: There is only one expression object for the integer value zero;
3044: it is the value of the variable @code{const0_rtx}. Likewise, the
3045: only expression for integer value one is found in @code{const1_rtx}.
1.1.1.2 root 3046: Any attempt to create an expression of code @samp{const_int} and
1.1 root 3047: value zero or one will return @code{const0_rtx} or @code{const1_rtx}
3048: as appropriate.
3049:
3050: @item (const_double:@var{m} @var{i0} @var{i1})
3051: Represents a floating point constant value of mode @var{m}. The two
3052: integers @var{i0} and @var{i1} together contain the bits of a
3053: @code{double} value. To convert them to a @code{double}, do
3054:
3055: @example
1.1.1.2 root 3056: union @{ double d; int i[2];@} u;
1.1 root 3057: u.i[0] = XINT (x, 0);
3058: u.i[1] = XINT (x, 1);
3059: @end example
3060:
3061: @noindent
3062: and then refer to @code{u.d}. The value of the constant is
3063: represented as a double in this fashion even if the value represented
3064: is single-precision.
3065:
1.1.1.2 root 3066: The global variables @code{dconst0_rtx} and @code{fconst0_rtx} hold
3067: @samp{const_double} expressions with value 0, in modes @code{DFmode} and
3068: @code{SFmode}, respectively.
1.1 root 3069:
3070: @item (symbol_ref @var{symbol})
3071: Represents the value of an assembler label for data. @var{symbol} is
3072: a string that describes the name of the assembler label. If it starts
3073: with a @samp{*}, the label is the rest of @var{symbol} not including
3074: the @samp{*}. Otherwise, the label is @var{symbol}, prefixed with
3075: @samp{_}.
3076:
3077: @item (label_ref @var{label})
3078: Represents the value of an assembler label for code. It contains one
1.1.1.2 root 3079: operand, an expression, which must be a @samp{code_label} that appears
1.1 root 3080: in the instruction sequence to identify the place where the label
3081: should go.
3082:
3083: The reason for using a distinct expression type for code label
3084: references is so that jump optimization can distinguish them.
3085:
3086: @item (const @var{exp})
3087: Represents a constant that is the result of an assembly-time
3088: arithmetic computation. The operand, @var{exp}, is an expression that
3089: contains only constants (@samp{const_int}, @samp{symbol_ref} and
3090: @samp{label_ref} expressions) combined with @samp{plus} and
3091: @samp{minus}. However, not all combinations are valid, since the
3092: assembler cannot do arbitrary arithmetic on relocatable symbols.
3093: @end table
3094:
3095: @node Regs and Memory, Arithmetic, Constants, RTL
3096: @section Registers and Memory
3097:
3098: Here are the RTL expression types for describing access to machine
3099: registers and to main memory.
3100:
3101: @table @code
3102: @item (reg:@var{m} @var{n})
3103: For small values of the integer @var{n} (less than
3104: @code{FIRST_PSEUDO_REGISTER}), this stands for a reference to machine
3105: register number @var{n}: a @dfn{hard register}. For larger values of
3106: @var{n}, it stands for a temporary value or @dfn{pseudo register}.
3107: The compiler's strategy is to generate code assuming an unlimited
3108: number of such pseudo registers, and later convert them into hard
3109: registers or into memory references.
3110:
3111: The symbol @code{FIRST_PSEUDO_REGISTER} is defined by the machine
3112: description, since the number of hard registers on the machine is an
3113: invariant characteristic of the machine. Note, however, that not
3114: all of the machine registers must be general registers. All the
3115: machine registers that can be used for storage of data are given
3116: hard register numbers, even those that can be used only in certain
3117: instructions or can hold only certain types of data.
3118:
1.1.1.2 root 3119: Each pseudo register number used in a function's RTL code is
1.1 root 3120: represented by a unique @samp{reg} expression.
3121:
3122: @var{m} is the machine mode of the reference. It is necessary because
3123: machines can generally refer to each register in more than one mode.
3124: For example, a register may contain a full word but there may be
3125: instructions to refer to it as a half word or as a single byte, as
3126: well as instructions to refer to it as a floating point number of
3127: various precisions.
3128:
3129: Even for a register that the machine can access in only one mode,
3130: the mode must always be specified.
3131:
3132: A hard register may be accessed in various modes throughout one
3133: function, but each pseudo register is given a natural mode
3134: and is accessed only in that mode. When it is necessary to describe
3135: an access to a pseudo register using a nonnatural mode, a @samp{subreg}
3136: expression is used.
3137:
3138: A @samp{reg} expression with a machine mode that specifies more than
3139: one word of data may actually stand for several consecutive registers.
3140: If in addition the register number specifies a hardware register, then
3141: it actually represents several consecutive hardware registers starting
3142: with the specified one.
3143:
3144: Such multi-word hardware register @samp{reg} expressions may not be live
3145: across the boundary of a basic block. The lifetime analysis pass does not
3146: know how to record properly that several consecutive registers are
3147: actually live there, and therefore register allocation would be confused.
3148: The CSE pass must go out of its way to make sure the situation does
3149: not arise.
3150:
3151: @item (subreg:@var{m} @var{reg} @var{wordnum})
3152: @samp{subreg} expressions are used to refer to a register in a machine
3153: mode other than its natural one, or to refer to one register of
3154: a multi-word @samp{reg} that actually refers to several registers.
3155:
3156: Each pseudo-register has a natural mode. If it is necessary to
3157: operate on it in a different mode---for example, to perform a fullword
3158: move instruction on a pseudo-register that contains a single byte---
3159: the pseudo-register must be enclosed in a @samp{subreg}. In such
3160: a case, @var{wordnum} is zero.
3161:
3162: The other use of @samp{subreg} is to extract the individual registers
3163: of a multi-register value. Machine modes such as @code{DImode} and
3164: @code{EPmode} indicate values longer than a word, values which usually
3165: require two consecutive registers. To access one of the registers,
3166: use a @samp{subreg} with mode @code{SImode} and a @var{wordnum} that
3167: says which register.
3168:
3169: The compilation parameter @code{WORDS_BIG_ENDIAN}, if defined, says
3170: that word number zero is the most significant part; otherwise, it is
3171: the least significant part.
3172:
3173: Note that it is not valid to access a @code{DFmode} value in @code{SFmode}
3174: using a @samp{subreg}. On some machines the most significant part of a
3175: @code{DFmode} value does not have the same format as a single-precision
3176: floating value.
3177:
3178: @item (cc0)
3179: This refers to the machine's condition code register. It has no
3180: operands and may not have a machine mode. It may be validly used in
3181: only two contexts: as the destination of an assignment (in test and
3182: compare instructions) and in comparison operators comparing against
1.1.1.2 root 3183: zero (@samp{const_int} with value zero; that is to say,
3184: @code{const0_rtx}).
1.1 root 3185:
1.1.1.2 root 3186: There is only one expression object of code @samp{cc0}; it is the
1.1 root 3187: value of the variable @code{cc0_rtx}. Any attempt to create an
1.1.1.2 root 3188: expression of code @samp{cc0} will return @code{cc0_rtx}.
1.1 root 3189:
1.1.1.2 root 3190: One special thing about the condition code register is that
3191: instructions can set it implicitly. On many machines, nearly all
3192: instructions set the condition code based on the value that they
3193: compute or store. It is not necessary to record these actions
3194: explicitly in the RTL because the machine description includes a
3195: prescription for recognizing the instructions that do so (by means of
3196: the macro @code{NOTICE_UPDATE_CC}). Only instructions whose sole
3197: purpose is to set the condition code, and instructions that use the
3198: condition code, need mention @code{(cc0)}.
1.1 root 3199:
3200: @item (pc)
3201: This represents the machine's program counter. It has no operands and
3202: may not have a machine mode. @code{(pc)} may be validly used only in
3203: certain specific contexts in jump instructions.
3204:
1.1.1.2 root 3205: There is only one expression object of code @samp{pc}; it is the value
3206: of the variable @code{pc_rtx}. Any attempt to create an expression of
3207: code @samp{pc} will return @code{pc_rtx}.
1.1 root 3208:
1.1.1.2 root 3209: All instructions that do not jump alter the program counter implicitly
3210: by incrementing it, but there is no need to mention this in the RTL.
1.1 root 3211:
3212: @item (mem:@var{m} @var{addr})
1.1.1.2 root 3213: This RTX represents a reference to main memory at an address
3214: represented by the expression @var{addr}. @var{m} specifies how large
3215: a unit of memory is accessed.
1.1 root 3216: @end table
3217:
3218: @node Arithmetic, Comparisons, Regs and Memory, RTL
3219: @section RTL Expressions for Arithmetic
3220:
3221: @table @code
3222: @item (plus:@var{m} @var{x} @var{y})
3223: Represents the sum of the values represented by @var{x} and @var{y}
3224: carried out in machine mode @var{m}. This is valid only if
3225: @var{x} and @var{y} both are valid for mode @var{m}.
3226:
3227: @item (minus:@var{m} @var{x} @var{y})
3228: Like @samp{plus} but represents subtraction.
3229:
3230: @item (minus @var{x} @var{y})
3231: Represents the result of subtracting @var{y} from @var{x}
3232: for purposes of comparison. The absence of a machine mode
3233: in the @samp{minus} expression indicates that the result is
3234: computed without overflow, as if with infinite precision.
3235:
3236: Of course, machines can't really subtract with infinite precision.
3237: However, they can pretend to do so when only the sign of the
3238: result will be used, which is the case when the result is stored
1.1.1.2 root 3239: in @code{(cc0)}. And that is the only way this kind of expression
1.1 root 3240: may validly be used: as a value to be stored in the condition codes.
3241:
3242: @item (neg:@var{m} @var{x})
3243: Represents the negation (subtraction from zero) of the value
3244: represented by @var{x}, carried out in mode @var{m}. @var{x} must be
3245: valid for mode @var{m}.
3246:
3247: @item (mult:@var{m} @var{x} @var{y})
3248: Represents the signed product of the values represented by @var{x} and
3249: @var{y} carried out in machine mode @var{m}. If
3250: @var{x} and @var{y} are both valid for mode @var{m}, this is ordinary
1.1.1.2 root 3251: size-preserving multiplication. Alternatively, both @var{x} and @var{y}
1.1 root 3252: may be valid for a different, narrower mode. This represents the
3253: kind of multiplication that generates a product wider than the operands.
3254: Widening multiplication and same-size multiplication are completely
3255: distinct and supported by different machine instructions; machines may
1.1.1.2 root 3256: support one but not the other.@refill
1.1 root 3257:
3258: @samp{mult} may be used for floating point division as well.
3259: Then @var{m} is a floating point machine mode.
3260:
3261: @item (umult:@var{m} @var{x} @var{y})
3262: Like @samp{mult} but represents unsigned multiplication. It may be
3263: used in both same-size and widening forms, like @samp{mult}.
1.1.1.2 root 3264: @samp{umult} is used only for fixed-point multiplication.
1.1 root 3265:
3266: @item (div:@var{m} @var{x} @var{y})
3267: Represents the quotient in signed division of @var{x} by @var{y},
3268: carried out in machine mode @var{m}. If @var{m} is a floating-point
3269: mode, it represents the exact quotient; otherwise, the integerized
3270: quotient. If @var{x} and @var{y} are both valid for mode @var{m},
3271: this is ordinary size-preserving division. Some machines have
3272: division instructions in which the operands and quotient widths are
3273: not all the same; such instructions are represented by @samp{div}
3274: expressions in which the machine modes are not all the same.
3275:
3276: @item (udiv:@var{m} @var{x} @var{y})
3277: Like @samp{div} but represents unsigned division.
3278:
3279: @item (mod:@var{m} @var{x} @var{y})
3280: @itemx (umod:@var{m} @var{x} @var{y})
3281: Like @samp{div} and @samp{udiv} but represent the remainder instead of
3282: the quotient.
3283:
3284: @item (not:@var{m} @var{x})
3285: Represents the bitwise complement of the value represented by @var{x},
3286: carried out in mode @var{m}, which must be a fixed-point machine mode.
3287: @var{x} must be valid for mode @var{m}, which must be a fixed-point mode.
3288:
3289: @item (and:@var{m} @var{x} @var{y})
3290: Represents the bitwise logical-and of the values represented by
3291: @var{x} and @var{y}, carried out in machine mode @var{m}. This is
3292: valid only if @var{x} and @var{y} both are valid for mode @var{m},
3293: which must be a fixed-point mode.
3294:
3295: @item (ior:@var{m} @var{x} @var{y})
3296: Represents the bitwise inclusive-or of the values represented by
3297: @var{x} and @var{y}, carried out in machine mode @var{m}. This is
3298: valid only if @var{x} and @var{y} both are valid for mode @var{m},
3299: which must be a fixed-point mode.
3300:
3301: @item (xor:@var{m} @var{x} @var{y})
3302: Represents the bitwise exclusive-or of the values represented by
3303: @var{x} and @var{y}, carried out in machine mode @var{m}. This is
3304: valid only if @var{x} and @var{y} both are valid for mode @var{m},
3305: which must be a fixed-point mode.
3306:
3307: @item (lshift:@var{m} @var{x} @var{c})
3308: Represents the result of logically shifting @var{x} left by @var{c}
3309: places. @var{x} must be valid for the mode @var{m}, a fixed-point
3310: machine mode. @var{c} must be valid for a fixed-point mode;
3311: which mode is determined by the mode called for in the machine
3312: description entry for the left-shift instruction. For example,
3313: on the Vax, the mode of @var{c} is @code{QImode} regardless of @var{m}.
3314:
3315: On some machines, negative values of @var{c} may be meaningful; this
1.1.1.2 root 3316: is why logical left shift and arithmetic left shift are distinguished.
1.1 root 3317: For example, Vaxes have no right-shift instructions, and right shifts
3318: are represented as left-shift instructions whose counts happen
3319: to be negative constants or else computed (in a previous instruction)
3320: by negation.
3321:
3322: @item (ashift:@var{m} @var{x} @var{c})
3323: Like @samp{lshift} but for arithmetic left shift.
3324:
3325: @item (lshiftrt:@var{m} @var{x} @var{c})
3326: @itemx (ashiftrt:@var{m} @var{x} @var{c})
3327: Like @samp{lshift} and @samp{ashift} but for right shift.
3328:
3329: @item (rotate:@var{m} @var{x} @var{c})
3330: @itemx (rotatert:@var{m} @var{x} @var{c})
3331: Similar but represent left and right rotate.
3332:
3333: @item (abs:@var{m} @var{x})
3334: Represents the absolute value of @var{x}, computed in mode @var{m}.
3335: @var{x} must be valid for @var{m}.
3336:
3337: @item (sqrt:@var{m} @var{x})
3338: Represents the square root of @var{x}, computed in mode @var{m}.
3339: @var{x} must be valid for @var{m}. Most often @var{m} will be
3340: a floating point mode.
1.1.1.2 root 3341:
3342: @item (ffs:@var{m} @var{x})
3343: Represents the one plus the index of the least significant 1-bit in
3344: @var{x}, represented as an integer of mode @var{m}. (The value is
3345: zero if @var{x} is zero.) The mode of @var{x} need not be @var{m};
3346: depending on the target machine, various mode combinations may be
3347: valid.
1.1 root 3348: @end table
3349:
3350: @node Comparisons, Bit Fields, Arithmetic, RTL
3351: @section Comparison Operations
3352:
3353: Comparison operators test a relation on two operands and are considered to
3354: represent the value 1 if the relation holds, or zero if it does not. The
3355: mode of the comparison is determined by the operands; they must both be
3356: valid for a common machine mode. A comparison with both operands constant
3357: would be invalid as the machine mode could not be deduced from it, but such
1.1.1.2 root 3358: a comparison should never exist in RTL due to constant folding.
1.1 root 3359:
3360: Inequality comparisons come in two flavors, signed and unsigned. Thus,
1.1.1.2 root 3361: there are distinct expression codes @samp{gt} and @samp{gtu} for signed and
1.1 root 3362: unsigned greater-than. These can produce different results for the same
3363: pair of integer values: for example, 1 is signed greater-than -1 but not
3364: unsigned greater-than, because -1 when regarded as unsigned is actually
1.1.1.2 root 3365: @code{0xffffffff} which is greater than 1.
1.1 root 3366:
3367: The signed comparisons are also used for floating point values. Floating
3368: point comparisons are distinguished by the machine modes of the operands.
3369:
3370: The comparison operators may be used to compare the condition codes
1.1.1.2 root 3371: @code{(cc0)} against zero, as in @code{(eq (cc0) (const_int 0))}. Such a
3372: construct actually refers to the result of the preceding instruction in
3373: which the condition codes were set. The above example stands for 1 if the
3374: condition codes were set to say ``zero'' or ``equal'', 0 otherwise.
3375: Although the same comparison operators are used for this as may be used in
3376: other contexts on actual data, no confusion can result since the machine
3377: description would never allow both kinds of uses in the same context.
1.1 root 3378:
3379: @table @code
3380: @item (eq @var{x} @var{y})
3381: 1 if the values represented by @var{x} and @var{y} are equal,
3382: otherwise 0.
3383:
3384: @item (ne @var{x} @var{y})
3385: 1 if the values represented by @var{x} and @var{y} are not equal,
3386: otherwise 0.
3387:
3388: @item (gt @var{x} @var{y})
3389: 1 if the @var{x} is greater than @var{y}. If they are fixed-point,
3390: the comparison is done in a signed sense.
3391:
3392: @item (gtu @var{x} @var{y})
3393: Like @samp{gt} but does unsigned comparison, on fixed-point numbers only.
3394:
3395: @item (lt @var{x} @var{y})
3396: @item (ltu @var{x} @var{y})
3397: Like @samp{gt} and @samp{gtu} but test for ``less than''.
3398:
3399: @item (ge @var{x} @var{y})
3400: @item (geu @var{x} @var{y})
3401: Like @samp{gt} and @samp{gtu} but test for ``greater than or equal''.
3402:
3403: @item (le @var{x} @var{y})
3404: @item (leu @var{x} @var{y})
3405: Like @samp{gt} and @samp{gtu} but test for ``less than or equal''.
3406:
3407: @item (if_then_else @var{cond} @var{then} @var{else})
3408: This is not a comparison operation but is listed here because it is
3409: always used in conjunction with a comparison operation. To be
3410: precise, @var{cond} is a comparison expression. This expression
3411: represents a choice, according to @var{cond}, between the value
3412: represented by @var{then} and the one represented by @var{else}.
3413:
3414: On most machines, @samp{if_then_else} expressions are valid only
3415: to express conditional jumps.
3416: @end table
3417:
3418: @node Bit Fields, Conversions, Comparisons, RTL
3419: @section Bit-fields
3420:
3421: Special expression codes exist to represent bit-field instructions.
1.1.1.2 root 3422: These types of expressions are lvalues in RTL; they may appear
1.1 root 3423: on the left side of a assignment, indicating insertion of a value
3424: into the specified bit field.
3425:
3426: @table @code
3427: @item (sign_extract:SI @var{loc} @var{size} @var{pos})
3428: This represents a reference to a sign-extended bit-field contained or
3429: starting in @var{loc} (a memory or register reference). The bit field
3430: is @var{size} bits wide and starts at bit @var{pos}. The compilation
1.1.1.2 root 3431: option @code{BITS_BIG_ENDIAN} says which end of the memory unit
1.1 root 3432: @var{pos} counts from.
3433:
3434: Which machine modes are valid for @var{loc} depends on the machine,
3435: but typically @var{loc} should be a single byte when in memory
3436: or a full word in a register.
3437:
1.1.1.2 root 3438: @item (zero_extract:SI @var{loc} @var{size} @var{pos})
1.1 root 3439: Like @samp{sign_extract} but refers to an unsigned or zero-extended
3440: bit field. The same sequence of bits are extracted, but they
3441: are filled to an entire word with zeros instead of by sign-extension.
3442: @end table
3443:
3444: @node Conversions, RTL Declarations, Bit Fields, RTL
3445: @section Conversions
3446:
3447: All conversions between machine modes must be represented by
3448: explicit conversion operations. For example, an expression
1.1.1.2 root 3449: which is the sum of a byte and a full word cannot be written as
1.1 root 3450: @code{(plus:SI (reg:QI 34) (reg:SI 80))} because the @samp{plus}
3451: operation requires two operands of the same machine mode.
3452: Therefore, the byte-sized operand is enclosed in a conversion
3453: operation, as in
3454:
3455: @example
3456: (plus:SI (sign_extend:SI (reg:QI 34)) (reg:SI 80))
3457: @end example
3458:
3459: The conversion operation is not a mere placeholder, because there
3460: may be more than one way of converting from a given starting mode
3461: to the desired final mode. The conversion operation code says how
3462: to do it.
3463:
3464: @table @code
3465: @item (sign_extend:@var{m} @var{x})
3466: Represents the result of sign-extending the value @var{x}
3467: to machine mode @var{m}. @var{m} must be a fixed-point mode
3468: and @var{x} a fixed-point value of a mode narrower than @var{m}.
3469:
3470: @item (zero_extend:@var{m} @var{x})
3471: Represents the result of zero-extending the value @var{x}
3472: to machine mode @var{m}. @var{m} must be a fixed-point mode
3473: and @var{x} a fixed-point value of a mode narrower than @var{m}.
3474:
3475: @item (float_extend:@var{m} @var{x})
3476: Represents the result of extending the value @var{x}
3477: to machine mode @var{m}. @var{m} must be a floating point mode
3478: and @var{x} a floating point value of a mode narrower than @var{m}.
3479:
3480: @item (truncate:@var{m} @var{x})
3481: Represents the result of truncating the value @var{x}
3482: to machine mode @var{m}. @var{m} must be a fixed-point mode
3483: and @var{x} a fixed-point value of a mode wider than @var{m}.
3484:
3485: @item (float_truncate:@var{m} @var{x})
3486: Represents the result of truncating the value @var{x}
3487: to machine mode @var{m}. @var{m} must be a floating point mode
3488: and @var{x} a floating point value of a mode wider than @var{m}.
3489:
3490: @item (float:@var{m} @var{x})
1.1.1.2 root 3491: Represents the result of converting fixed point value @var{x},
3492: regarded as signed, to floating point mode @var{m}.
3493:
3494: @item (unsigned_float:@var{m} @var{x})
3495: Represents the result of converting fixed point value @var{x},
3496: regarded as unsigned, to floating point mode @var{m}.
1.1 root 3497:
3498: @item (fix:@var{m} @var{x})
1.1.1.2 root 3499: When @var{m} is a fixed point mode, represents the result of
3500: converting floating point value @var{x} to mode @var{m}, regarded as
3501: signed. How rounding is done is not specified, so this operation may
3502: be used validly in compiling C code only for integer-valued operands.
3503:
3504: @item (unsigned_fix:@var{m} @var{x})
3505: Represents the result of converting floating point value @var{x} to
3506: fixed point mode @var{m}, regarded as unsigned. How rounding is done
3507: is not specified.
1.1 root 3508:
1.1.1.2 root 3509: @item (fix:@var{m} @var{x})
3510: When @var{m} is a floating point mode, represents the result of
3511: converting floating point value @var{x} (valid for mode @var{m}) to an
3512: integer, still represented in floating point mode @var{m}, by rounding
3513: towards zero.
1.1 root 3514: @end table
3515:
3516: @node RTL Declarations, Side Effects, Conversions, RTL
3517: @section Declarations
3518:
3519: Declaration expression codes do not represent arithmetic operations
3520: but rather state assertions about their operands.
3521:
3522: @table @code
3523: @item (strict_low_part (subreg:@var{m} (reg:@var{n} @var{r}) 0))
3524: This expression code is used in only one context: operand 0 of a
3525: @samp{set} expression. In addition, the operand of this expression
3526: must be a @samp{subreg} expression.
3527:
3528: The presence of @samp{strict_low_part} says that the part of the
1.1.1.2 root 3529: register which is meaningful in mode @var{n}, but is not part of
3530: mode @var{m}, is not to be altered. Normally, an assignment to such
1.1 root 3531: a subreg is allowed to have undefined effects on the rest of the
3532: register when @var{m} is less than a word.
3533: @end table
3534:
3535: @node Side Effects, Incdec, RTL Declarations, RTL
3536: @section Side Effect Expressions
3537:
3538: The expression codes described so far represent values, not actions.
3539: But machine instructions never produce values; they are meaningful
3540: only for their side effects on the state of the machine. Special
3541: expression codes are used to represent side effects.
3542:
3543: The body of an instruction is always one of these side effect codes;
3544: the codes described above, which represent values, appear only as
3545: the operands of these.
3546:
3547: @table @code
3548: @item (set @var{lval} @var{x})
3549: Represents the action of storing the value of @var{x} into the place
3550: represented by @var{lval}. @var{lval} must be an expression
3551: representing a place that can be stored in: @samp{reg} (or
3552: @samp{subreg} or @samp{strict_low_part}), @samp{mem}, @samp{pc} or
1.1.1.2 root 3553: @samp{cc0}.@refill
1.1 root 3554:
3555: If @var{lval} is a @samp{reg}, @samp{subreg} or @samp{mem}, it has a
1.1.1.2 root 3556: machine mode; then @var{x} must be valid for that mode.@refill
1.1 root 3557:
3558: If @var{lval} is a @samp{reg} whose machine mode is less than the full
3559: width of the register, then it means that the part of the register
3560: specified by the machine mode is given the specified value and the
3561: rest of the register receives an undefined value. Likewise, if
3562: @var{lval} is a @samp{subreg} whose machine mode is narrower than
3563: @code{SImode}, the rest of the register can be changed in an undefined way.
3564:
3565: If @var{lval} is a @samp{strict_low_part} of a @samp{subreg}, then the
3566: part of the register specified by the machine mode of the
3567: @samp{subreg} is given the value @var{x} and the rest of the register
1.1.1.2 root 3568: is not changed.@refill
1.1 root 3569:
3570: If @var{lval} is @code{(cc0)}, it has no machine mode, and @var{x} may
1.1.1.2 root 3571: have any mode. This represents a ``test'' or ``compare'' instruction.@refill
1.1 root 3572:
3573: If @var{lval} is @code{(pc)}, we have a jump instruction, and the
3574: possibilities for @var{x} are very limited. It may be a
3575: @samp{label_ref} expression (unconditional jump). It may be an
3576: @samp{if_then_else} (conditional jump), in which case either the
3577: second or the third operand must be @code{(pc)} (for the case which
3578: does not jump) and the other of the two must be a @samp{label_ref}
3579: (for the case which does jump). @var{x} may also be a @samp{mem} or
3580: @code{(plus:SI (pc) @var{y})}, where @var{y} may be a @samp{reg} or a
3581: @samp{mem}; these unusual patterns are used to represent jumps through
1.1.1.2 root 3582: branch tables.@refill
1.1 root 3583:
3584: @item (return)
1.1.1.2 root 3585: Represents a return from the current function, on machines where this
3586: can be done with one instruction, such as Vaxes. On machines where a
3587: multi-instruction ``epilogue'' must be executed in order to return
3588: from the function, returning is done by jumping to a label which
3589: precedes the epilogue, and the @samp{return} expression code is never
3590: used.
1.1 root 3591:
3592: @item (call @var{function} @var{nargs})
3593: Represents a function call. @var{function} is a @samp{mem} expression
3594: whose address is the address of the function to be called. @var{nargs}
3595: is an expression representing the number of words of argument.
3596:
3597: Each machine has a standard machine mode which @var{function} must
1.1.1.2 root 3598: have. The machine description defines macro @code{FUNCTION_MODE} to
1.1 root 3599: expand into the requisite mode name. The purpose of this mode is to
3600: specify what kind of addressing is allowed, on machines where the
3601: allowed kinds of addressing depend on the machine mode being
3602: addressed.
3603:
3604: @item (clobber @var{x})
3605: Represents the storing or possible storing of an unpredictable,
3606: undescribed value into @var{x}, which must be a @samp{reg} or
3607: @samp{mem} expression.
3608:
3609: One place this is used is in string instructions that store standard
3610: values into particular hard registers. It may not be worth the
1.1.1.2 root 3611: trouble to describe the values that are stored, but it is essential to
3612: inform the compiler that the registers will be altered, lest it
1.1 root 3613: attempt to keep data in them across the string instruction.
3614:
3615: @var{x} may also be null---a null C pointer, no expression at all.
3616: Such a @code{(clobber (null))} expression means that all memory
3617: locations must be presumed clobbered.
3618:
3619: Note that the machine description classifies certain hard registers as
3620: ``call-clobbered''. All function call instructions are assumed by
3621: default to clobber these registers, so there is no need to use
3622: @samp{clobber} expressions to indicate this fact. Also, each function
3623: call is assumed to have the potential to alter any memory location.
3624:
3625: @item (use @var{x})
1.1.1.2 root 3626: Represents the use of the value of @var{x}. It indicates that the
3627: value in @var{x} at this point in the program is needed, even though
3628: it may not be apparent why this is so. Therefore, the compiler will
3629: not attempt to delete instructions whose only effect is to store a
3630: value in @var{x}. @var{x} must be a @samp{reg} expression.
1.1 root 3631:
3632: @item (parallel [@var{x0} @var{x1} @dots{}])
3633: Represents several side effects performed in parallel. The square
3634: brackets stand for a vector; the operand of @samp{parallel} is a
3635: vector of expressions. @var{x0}, @var{x1} and so on are individual
3636: side effects---expressions of code @samp{set}, @samp{call},
1.1.1.2 root 3637: @samp{return}, @samp{clobber} or @samp{use}.@refill
1.1 root 3638:
1.1.1.2 root 3639: ``In parallel'' means that first all the values used in the individual
3640: side-effects are computed, and second all the actual side-effects are
3641: performed. For example,
1.1 root 3642:
3643: @example
3644: (parallel [(set (reg:SI 1) (mem:SI (reg:SI 1)))
3645: (set (mem:SI (reg:SI 1)) (reg:SI 1))])
3646: @end example
3647:
3648: @noindent
3649: says unambiguously that the values of hard register 1 and the memory
3650: location addressed by it are interchanged. In both places where
3651: @code{(reg:SI 1)} appears as a memory address it refers to the value
1.1.1.2 root 3652: in register 1 @emph{before} the execution of the instruction.
3653:
3654: @item (sequence [@var{insns} @dots{}])
3655: Represents a sequence of insns. Each of the @var{insns} that appears
3656: in the vector is suitable for appearing in the chain of insns, so it
3657: must be an @samp{insn}, @samp{jump_insn}, @samp{call_insn},
3658: @samp{code_label}, @samp{barrier} or @samp{note}.
3659:
3660: A @samp{sequence} RTX never appears in an actual insn. It represents
3661: the sequence of insns that result from a @samp{define_expand}
3662: @emph{before} those insns are passed to @code{emit_insn} to insert
3663: them in the chain of insns. When actually inserted, the individual
3664: sub-insns are separated out and the @samp{sequence} is forgotten.
1.1 root 3665: @end table
3666:
1.1.1.2 root 3667: Three expression codes appear in place of a side effect, as the body of an
3668: insn, though strictly speaking they do not describe side effects as such:
1.1 root 3669:
3670: @table @code
3671: @item (asm_input @var{s})
3672: Represents literal assembler code as described by the string @var{s}.
3673:
3674: @item (addr_vec:@var{m} [@var{lr0} @var{lr1} @dots{}])
1.1.1.2 root 3675: Represents a table of jump addresses. The vector elements @var{lr0},
3676: etc., are @samp{label_ref} expressions. The mode @var{m} specifies
3677: how much space is given to each address; normally @var{m} would be
1.1 root 3678: @code{Pmode}.
3679:
3680: @item (addr_diff_vec:@var{m} @var{base} [@var{lr0} @var{lr1} @dots{}])
3681: Represents a table of jump addresses expressed as offsets from
1.1.1.2 root 3682: @var{base}. The vector elements @var{lr0}, etc., are @samp{label_ref}
3683: expressions and so is @var{base}. The mode @var{m} specifies how much
3684: space is given to each address-difference.@refill
1.1 root 3685: @end table
3686:
1.1.1.2 root 3687: @node Incdec, Assembler, Side Effects, RTL
1.1 root 3688: @section Embedded Side-Effects on Addresses
3689:
3690: Four special side-effect expression codes appear as memory addresses.
3691:
3692: @table @code
3693: @item (pre_dec:@var{m} @var{x})
3694: Represents the side effect of decrementing @var{x} by a standard
3695: amount and represents also the value that @var{x} has after being
3696: decremented. @var{x} must be a @samp{reg} or @samp{mem}, but most
3697: machines allow only a @samp{reg}. @var{m} must be the machine mode
1.1.1.2 root 3698: for pointers on the machine in use. The amount @var{x} is decremented
1.1 root 3699: by is the length in bytes of the machine mode of the containing memory
3700: reference of which this expression serves as the address. Here is an
1.1.1.2 root 3701: example of its use:@refill
1.1 root 3702:
3703: @example
3704: (mem:DF (pre_dec:SI (reg:SI 39)))
3705: @end example
3706:
3707: @noindent
3708: This says to decrement pseudo register 39 by the length of a @code{DFmode}
3709: value and use the result to address a @code{DFmode} value.
3710:
3711: @item (pre_inc:@var{m} @var{x})
3712: Similar, but specifies incrementing @var{x} instead of decrementing it.
3713:
3714: @item (post_dec:@var{m} @var{x})
3715: Represents the same side effect as @samp{pre_decrement} but a different
3716: value. The value represented here is the value @var{x} has @i{before}
3717: being decremented.
3718:
3719: @item (post_inc:@var{m} @var{x})
3720: Similar, but specifies incrementing @var{x} instead of decrementing it.
3721: @end table
3722:
3723: These embedded side effect expressions must be used with care. Instruction
3724: patterns may not use them. Until the @samp{flow} pass of the compiler,
3725: they may occur only to represent pushes onto the stack. The @samp{flow}
3726: pass finds cases where registers are incremented or decremented in one
3727: instruction and used as an address shortly before or after; these cases are
3728: then transformed to use pre- or post-increment or -decrement.
3729:
3730: Explicit popping of the stack could be represented with these embedded
3731: side effect operators, but that would not be safe; the instruction
3732: combination pass could move the popping past pushes, thus changing
3733: the meaning of the code.
3734:
3735: An instruction that can be represented with an embedded side effect
3736: could also be represented using @samp{parallel} containing an additional
3737: @samp{set} to describe how the address register is altered. This is not
3738: done because machines that allow these operations at all typically
3739: allow them wherever a memory address is called for. Describing them as
3740: additional parallel stores would require doubling the number of entries
3741: in the machine description.
3742:
1.1.1.2 root 3743: @node Assembler, Insns, IncDec, RTL
3744: @section Assembler Instructions as Expressions
3745:
3746: The RTX code @samp{asm_operands} represents a value produced by a
3747: user-specified assembler instruction. It is used to represent
3748: an @code{asm} statement with arguments. An @code{asm} statement with
3749: a single output operand, like this:
3750:
3751: @example
3752: asm ("foo %1,%2,%0" : "a" (outputvar) : "g" (x + y), "di" (*z));
3753: @end example
3754:
3755: @noindent
3756: is represented using a single @samp{asm_operands} RTX which represents
3757: the value that is stored in @code{outputvar}:
3758:
3759: @example
3760: (set @var{rtx-for-outputvar}
3761: (asm_operands "foo %1,%2,%0" "a" 0
3762: [@var{rtx-for-addition-result} @var{rtx-for-*z}]
3763: [(asm_input:@var{m1} "g")
3764: (asm_input:@var{m2} "di")]))
3765: @end example
3766:
3767: @noindent
3768: Here the operands of the @samp{asm_operands} RTX are the assembler
3769: template string, the output-operand's constraint, the index-number of the
3770: output operand among the output operands specified, a vector of input
3771: operand RTX's, and a vector of input-operand modes and constraints. The
3772: mode @var{m1} is the mode of the sum @code{x+y}; @var{m2} is that of
3773: @code{*z}.
3774:
3775: When an @code{asm} statement has multiple output values, its insn has
3776: several such @samp{set} RTX's inside of a @samp{parallel}. Each @samp{set}
3777: contains a @samp{asm_operands}; all of these share the same assembler
3778: template and vectors, but each contains the constraint for the respective
3779: output operand. They are also distinguished by the output-operand index
3780: number, which is 0, 1, @dots{} for successive output operands.
3781:
3782: @node Insns, Calls, Assembler, RTL
1.1 root 3783: @section Insns
3784:
3785: The RTL representation of the code for a function is a doubly-linked
3786: chain of objects called @dfn{insns}. Insns are expressions with
3787: special codes that are used for no other purpose. Some insns are
3788: actual instructions; others represent dispatch tables for @code{switch}
3789: statements; others represent labels to jump to or various sorts of
1.1.1.2 root 3790: declarative information.
1.1 root 3791:
1.1.1.2 root 3792: In addition to its own specific data, each insn must have a unique id-number
1.1 root 3793: that distinguishes it from all other insns in the current function, and
3794: chain pointers to the preceding and following insns. These three fields
3795: occupy the same position in every insn, independent of the expression code
3796: of the insn. They could be accessed with @code{XEXP} and @code{XINT},
3797: but instead three special macros are always used:
3798:
3799: @table @code
3800: @item INSN_UID (@var{i})
3801: Accesses the unique id of insn @var{i}.
3802:
3803: @item PREV_INSN (@var{i})
3804: Accesses the chain pointer to the insn preceding @var{i}.
3805: If @var{i} is the first insn, this is a null pointer.
3806:
3807: @item NEXT_INSN (@var{i})
3808: Accesses the chain pointer to the insn following @var{i}.
3809: If @var{i} is the last insn, this is a null pointer.
3810: @end table
3811:
3812: The @code{NEXT_INSN} and @code{PREV_INSN} pointers must always
3813: correspond: if @var{i} is not the first insn,
3814:
3815: @example
3816: NEXT_INSN (PREV_INSN (@var{insn})) == @var{insn}
3817: @end example
3818:
3819: @noindent
3820: is always true.
3821:
3822: Every insn has one of the following six expression codes:
3823:
1.1.1.2 root 3824: @table @samp
1.1 root 3825: @item insn
3826: The expression code @samp{insn} is used for instructions that do not jump
3827: and do not do function calls. Insns with code @samp{insn} have four
3828: additional fields beyond the three mandatory ones listed above.
3829: These four are described in a table below.
3830:
3831: @item jump_insn
3832: The expression code @samp{jump_insn} is used for instructions that may jump
3833: (or, more generally, may contain @samp{label_ref} expressions).
3834: @samp{jump_insn} insns have the same extra fields as @samp{insn} insns,
3835: accessed in the same way.
3836:
3837: @item call_insn
3838: The expression code @samp{call_insn} is used for instructions that may do
3839: function calls. It is important to distinguish these instructions because
3840: they imply that certain registers and memory locations may be altered
3841: unpredictably.
3842:
3843: @samp{call_insn} insns have the same extra fields as @samp{insn} insns,
3844: accessed in the same way.
3845:
3846: @item code_label
3847: A @samp{code_label} insn represents a label that a jump insn can jump to.
3848: It contains one special field of data in addition to the three standard ones.
3849: It is used to hold the @dfn{label number}, a number that identifies this
3850: label uniquely among all the labels in the compilation (not just in the
3851: current function). Ultimately, the label is represented in the assembler
3852: output as an assembler label @samp{L@var{n}} where @var{n} is the label number.
3853:
3854: @item barrier
3855: Barriers are placed in the instruction stream after unconditional
3856: jump instructions to indicate that the jumps are unconditional.
3857: They contain no information beyond the three standard fields.
3858:
3859: @item note
3860: @samp{note} insns are used to represent additional debugging and
1.1.1.2 root 3861: declarative information. They contain two nonstandard fields, an
1.1 root 3862: integer which is accessed with the macro @code{NOTE_LINE_NUMBER} and a
3863: string accessed with @code{NOTE_SOURCE_FILE}.
3864:
3865: If @code{NOTE_LINE_NUMBER} is positive, the note represents the
3866: position of a source line and @code{NOTE_SOURCE_FILE} is the source file name
3867: that the line came from. These notes control generation of line
3868: number data in the assembler output.
3869:
3870: Otherwise, @code{NOTE_LINE_NUMBER} is not really a line number but a
3871: code with one of the following values (and @code{NOTE_SOURCE_FILE}
3872: must contain a null pointer):
3873:
3874: @table @code
3875: @item NOTE_INSN_DELETED
3876: Such a note is completely ignorable. Some passes of the compiler
3877: delete insns by altering them into notes of this kind.
3878:
3879: @item NOTE_INSN_BLOCK_BEG
3880: @itemx NOTE_INSN_BLOCK_END
3881: These types of notes indicate the position of the beginning and end
3882: of a level of scoping of variable names. They control the output
3883: of debugging information.
3884:
3885: @item NOTE_INSN_LOOP_BEG
3886: @itemx NOTE_INSN_LOOP_END
3887: These types of notes indicate the position of the beginning and end
3888: of a @code{while} or @code{for} loop. They enable the loop optimizer
3889: to find loops quickly.
3890: @end table
3891: @end table
3892:
3893: Here is a table of the extra fields of @samp{insn}, @samp{jump_insn}
3894: and @samp{call_insn} insns:
3895:
3896: @table @code
3897: @item PATTERN (@var{i})
3898: An expression for the side effect performed by this insn.
3899:
3900: @item REG_NOTES (@var{i})
3901: A list (chain of @samp{expr_list} expressions) giving information
3902: about the usage of registers in this insn. This list is set up by the
1.1.1.2 root 3903: flow analysis pass; it is a null pointer until then.
1.1 root 3904:
3905: @item LOG_LINKS (@var{i})
3906: A list (chain of @samp{insn_list} expressions) of previous ``related''
3907: insns: insns which store into registers values that are used for the
3908: first time in this insn. (An additional constraint is that neither a
3909: jump nor a label may come between the related insns). This list is
1.1.1.2 root 3910: set up by the flow analysis pass; it is a null pointer until then.
1.1 root 3911:
3912: @item INSN_CODE (@var{i})
3913: An integer that says which pattern in the machine description matches
3914: this insn, or -1 if the matching has not yet been attempted.
3915:
3916: Such matching is never attempted and this field is not used on an insn
3917: whose pattern consists of a single @samp{use}, @samp{clobber},
3918: @samp{asm}, @samp{addr_vec} or @samp{addr_diff_vec} expression.
3919: @end table
3920:
3921: The @code{LOG_LINKS} field of an insn is a chain of @samp{insn_list}
3922: expressions. Each of these has two operands: the first is an insn,
3923: and the second is another @samp{insn_list} expression (the next one in
3924: the chain). The last @samp{insn_list} in the chain has a null pointer
3925: as second operand. The significant thing about the chain is which
1.1.1.2 root 3926: insns appear in it (as first operands of @samp{insn_list}
1.1 root 3927: expressions). Their order is not significant.
3928:
3929: The @code{REG_NOTES} field of an insn is a similar chain but of
1.1.1.2 root 3930: @samp{expr_list} expressions instead of @samp{insn_list}. There are four
3931: kinds of register notes, which are distinguished by the machine mode of the
3932: @samp{expr_list}, which a register note is really understood as being an
3933: @code{enum reg_note}. The first operand @var{op} of the @samp{expr_list}
3934: is data whose meaning depends on the kind of note. Here are the four
3935: kinds:
1.1 root 3936:
3937: @table @code
3938: @item REG_DEAD
1.1.1.2 root 3939: The register @var{op} dies in this insn; that is to say, altering the
3940: value immediately after this insn would not affect the future behavior
3941: of the program.
1.1 root 3942:
3943: @item REG_INC
1.1.1.2 root 3944: The register @var{op} is incremented (or decremented; at this level
1.1 root 3945: there is no distinction) by an embedded side effect inside this insn.
1.1.1.2 root 3946: This means it appears in a @code{POST_INC}, @code{PRE_INC},
3947: @code{POST_DEC} or @code{PRE_DEC} RTX.
1.1 root 3948:
1.1.1.2 root 3949: @item REG_EQUIV
3950: The register that is set by this insn will be equal to @var{op} at run
3951: time, and could validly be replaced in all its occurrences by
3952: @var{op}. (``Validly'' here refers to the data flow of the program;
3953: simple replacement may make some insns invalid.)
3954:
3955: The value which the insn explicitly copies into the register may look
3956: different from @var{op}, but they will be equal at run time.
3957:
3958: For example, when a constant is loaded into a register that is never
3959: assigned any other value, this kind of note is used.
3960:
3961: When a parameter is copied into a pseudo-register at entry to a function,
3962: a note of this kind records that the register is equivalent to the stack
3963: slot where the parameter was passed. Although in this case the register
3964: may be set by other insns, it is still valid to replace the register
3965: by the stack slot throughout the function.
3966:
3967: @item REG_EQUAL
3968: The register that is set by this insn will be equal to @var{op} at run
3969: time at the end of this insn (but not necessarily elsewhere in the
3970: function).
3971:
3972: The RTX @var{op} is typically an arithmetic expression. For example,
3973: when a sequence of insns such as a library call is used to perform an
3974: arithmetic operation, this kind of note is attached to the insn that
3975: produces or copies the final value. It tells the CSE pass how to
3976: think of that value.
3977:
3978: @item REG_RETVAL
3979: This insn copies the value of a library call, and @var{op} is the
3980: first insn that was generated to set up the arguments for the library
3981: call.
3982:
3983: Flow analysis uses this note to delete all of a library call whose
3984: result is dead.
1.1 root 3985:
3986: @item REG_WAS_0
1.1.1.2 root 3987: The register @var{op} contained zero before this insn. You can rely
1.1 root 3988: on this note if it is present; its absence implies nothing.
3989: @end table
3990:
3991: (The only difference between the expression codes @samp{insn_list} and
3992: @samp{expr_list} is that the first operand of an @samp{insn_list} is
3993: assumed to be an insn and is printed in debugging dumps as the insn's
3994: unique id; the first operand of an @samp{expr_list} is printed in the
3995: ordinary way as an expression.)
3996:
1.1.1.2 root 3997: @node Calls, Sharing, Insns, RTL
3998: @section RTL Representation of Function-Call Insns
3999:
4000: Insns that call subroutines have the RTL expression code @samp{call_insn}.
4001: These insns must satisfy special rules, and their bodies must use a special
4002: RTL expression code, @samp{call}.
4003:
4004: A @samp{call} expression has two operands, as follows:
4005:
4006: @example
4007: (call @var{nbytes} (mem:@var{fm} @var{addr}))
4008: @end example
4009:
4010: @noindent
4011: Here @var{nbytes} is an operand that represents the number of bytes of
4012: argument data being passed to the subroutine, @var{fm} is a machine mode
4013: (which must equal as the definition of the @code{FUNCTION_MODE} macro in
4014: the machine description) and @var{addr} represents the address of the
4015: subroutine.
4016:
4017: For a subroutine that returns no value, the @samp{call} RTX as shown above
4018: is the entire body of the insn.
4019:
4020: For a subroutine that returns a value whose mode is not @code{BLKmode},
4021: the value is returned in a hard register. If this register's number is
4022: @var{r}, then the body of the call insn looks like this:
4023:
4024: @example
4025: (set (reg:@var{m} @var{r})
4026: (call @var{nbytes} (mem:@var{fm} @var{addr})))
4027: @end example
4028:
4029: @noindent
4030: This RTL expression makes it clear (to the optimizer passes) that the
4031: appropriate register receives a useful value in this insn.
4032:
4033: Immediately after RTL generation, if the value of the subroutine is
4034: actually used, this call insn is always followed closely by an insn which
1.1.1.3 root 4035: refers to the register @var{r}. This remains true through all the
4036: optimizer passes until cross jumping occurs.
4037:
4038: The following insn has one of two forms. Either it copies the value into a
4039: pseudo-register, like this:
1.1.1.2 root 4040:
4041: @example
4042: (set (reg:@var{m} @var{p}) (reg:@var{m} @var{r}))
4043: @end example
4044:
4045: @noindent
4046: or (in the case where the calling function will simply return whatever
4047: value the call produced, and no operation is needed to do this):
4048:
4049: @example
4050: (use (reg:@var{m} @var{r}))
4051: @end example
4052:
4053: @noindent
1.1.1.3 root 4054: Between the call insn and this following insn there may intervene only a
1.1.1.2 root 4055: stack-adjustment insn (and perhaps some @samp{note} insns).
4056:
4057: When a subroutine returns a @code{BLKmode} value, it is handled by
4058: passing to the subroutine the address of a place to store the value.
4059: So the call insn itself does not ``return'' any value, and it has the
4060: same RTL form as a call that returns nothing.
4061:
4062: @node Sharing,, Calls, RTL
1.1 root 4063: @section Structure Sharing Assumptions
4064:
4065: The compiler assumes that certain kinds of RTL expressions are unique;
4066: there do not exist two distinct objects representing the same value.
4067: In other cases, it makes an opposite assumption: that no RTL expression
4068: object of a certain kind appears in more than one place in the
4069: containing structure.
4070:
4071: These assumptions refer to a single function; except for the RTL
4072: objects that describe global variables and external functions,
4073: no RTL objects are common to two functions.
4074:
4075: @itemize @bullet
4076: @item
4077: Each pseudo-register has only a single @samp{reg} object to represent it,
4078: and therefore only a single machine mode.
4079:
4080: @item
4081: For any symbolic label, there is only one @samp{symbol_ref} object
4082: referring to it.
4083:
4084: @item
4085: There is only one @samp{const_int} expression with value zero,
4086: and only one with value one.
4087:
4088: @item
4089: There is only one @samp{pc} expression.
4090:
4091: @item
4092: There is only one @samp{cc0} expression.
4093:
4094: @item
4095: There is only one @samp{const_double} expression with mode
4096: @code{SFmode} and value zero, and only one with mode @code{DFmode} and
4097: value zero.
4098:
4099: @item
1.1.1.2 root 4100: No @samp{label_ref} appears in more than one place in the RTL
4101: structure; in other words, it is safe to do a tree-walk of all the
4102: insns in the function and assume that each time a @samp{label_ref} is
4103: seen it is distinct from all others that are seen.
1.1 root 4104:
4105: @item
1.1.1.2 root 4106: Only one @samp{mem} object is normally created for each static
4107: variable or stack slot, so these objects are frequently shared in all
4108: the places they appear. However, separate but equal objects for these
4109: variables are occasionally made.
4110:
4111: @item
4112: No RTL object appears in more than one place in the RTL structure
4113: except as described above. Many passes of the compiler rely on this
4114: by assuming that they can modify RTL objects in place without unwanted
4115: side-effects on other insns.
4116:
4117: @item
4118: During initial RTL generation, shared structure is freely introduced.
4119: After all the RTL for a function has been generated, all shared
4120: structure is copied by @code{unshare_all_rtl} in @file{emit-rtl.c},
4121: after which the above rules are guaranteed to be followed.
4122:
4123: @item
4124: During the combiner pass, shared structure with an insn can exist
4125: temporarily. However, the shared structure is copied before the
4126: combiner is finished with the insn. This is done by
4127: @code{copy_substitutions} in @samp{combine.c}.
1.1 root 4128: @end itemize
4129:
4130: @node Machine Desc, Machine Macros, RTL, Top
4131: @chapter Machine Descriptions
4132:
4133: A machine description has two parts: a file of instruction patterns
4134: (@file{.md} file) and a C header file of macro definitions.
4135:
4136: The @file{.md} file for a target machine contains a pattern for each
4137: instruction that the target machine supports (or at least each instruction
4138: that is worth telling the compiler about). It may also contain comments.
4139: A semicolon causes the rest of the line to be a comment, unless the semicolon
4140: is inside a quoted string.
4141:
4142: See the next chapter for information on the C header file.
4143:
4144: @menu
4145: * Patterns:: How to write instruction patterns.
1.1.1.2 root 4146: * Example:: An explained example of a @samp{define_insn} pattern.
4147: * RTL Template:: The RTL template defines what insns match a pattern.
4148: * Output Template:: The output template says how to make assembler code
4149: from such an insn.
4150: * Output Statement:: For more generality, write C code to output
4151: the assembler code.
1.1 root 4152: * Constraints:: When not all operands are general operands.
4153: * Standard Names:: Names mark patterns to use for code generation.
1.1.1.2 root 4154: * Pattern Ordering:: When the order of patterns makes a difference.
1.1 root 4155: * Dependent Patterns:: Having one pattern may make you need another.
1.1.1.2 root 4156: * Jump Patterns:: Special considerations for patterns for jump insns.
4157: * Peephole Definitions::Defining machine-specific peephole optimizations.
4158: * Expander Definitions::Generating a sequence of several RTL insns
4159: for a standard operation.
1.1 root 4160: @end menu
4161:
4162: @node Patterns, Example, Machine Desc, Machine Desc
1.1.1.2 root 4163: @section Everything about Instruction Patterns
1.1 root 4164:
4165: Each instruction pattern contains an incomplete RTL expression, with pieces
4166: to be filled in later, operand constraints that restrict how the pieces can
4167: be filled in, and an output pattern or C code to generate the assembler
4168: output, all wrapped up in a @samp{define_insn} expression.
4169:
1.1.1.2 root 4170: A @samp{define_insn} is an RTL expression containing four operands:
1.1 root 4171:
4172: @enumerate
4173: @item
4174: An optional name. The presence of a name indicate that this instruction
4175: pattern can perform a certain standard job for the RTL-generation
4176: pass of the compiler. This pass knows certain names and will use
4177: the instruction patterns with those names, if the names are defined
4178: in the machine description.
4179:
4180: The absence of a name is indicated by writing an empty string
4181: where the name should go. Nameless instruction patterns are never
4182: used for generating RTL code, but they may permit several simpler insns
4183: to be combined later on.
4184:
4185: Names that are not thus known and used in RTL-generation have no
4186: effect; they are equivalent to no name at all.
4187:
4188: @item
1.1.1.2 root 4189: The @dfn{RTL template} (@pxref{RTL Template}) is a vector of
4190: incomplete RTL expressions which show what the instruction should look
4191: like. It is incomplete because it may contain @samp{match_operand}
4192: and @samp{match_dup} expressions that stand for operands of the
1.1 root 4193: instruction.
4194:
4195: If the vector has only one element, that element is what the
4196: instruction should look like. If the vector has multiple elements,
4197: then the instruction looks like a @samp{parallel} expression
4198: containing that many elements as described.
4199:
4200: @item
4201: A condition. This is a string which contains a C expression that is
4202: the final test to decide whether an insn body matches this pattern.
4203:
4204: For a named pattern, the condition (if present) may not depend on
4205: the data in the insn being matched, but only the target-machine-type
4206: flags. The compiler needs to test these conditions during
4207: initialization in order to learn exactly which named instructions are
4208: available in a particular run.
4209:
4210: For nameless patterns, the condition is applied only when matching an
4211: individual insn, and only after the insn has matched the pattern's
4212: recognition template. The insn's operands may be found in the vector
4213: @code{operands}.
4214:
4215: @item
1.1.1.2 root 4216: The @dfn{output template}: a string that says how to output matching
4217: insns as assembler code. @samp{%} in this string specifies where
4218: to substitute the value of an operand. @xref{Output Template}.
4219:
4220: When simple substitution isn't general enough, you can specify a piece
4221: of C code to compute the output. @xref{Output Statement}.
4222: @end enumerate
4223:
4224: @node Example, RTL Template, Patterns, Machine Desc
4225: @section Example of @samp{define_insn}
4226:
4227: Here is an actual example of an instruction pattern, for the 68000/68020.
4228:
4229: @example
4230: (define_insn "tstsi"
4231: [(set (cc0)
4232: (match_operand:SI 0 "general_operand" "rm"))]
4233: ""
4234: "*
4235: @{ if (TARGET_68020 || ! ADDRESS_REG_P (operands[0]))
4236: return \"tstl %0\";
4237: return \"cmpl #0,%0\"; @}")
4238: @end example
1.1 root 4239:
1.1.1.2 root 4240: This is an instruction that sets the condition codes based on the value of
4241: a general operand. It has no condition, so any insn whose RTL description
4242: has the form shown may be handled according to this pattern. The name
4243: @samp{tstsi} means ``test a @code{SImode} value'' and tells the RTL generation
4244: pass that, when it is necessary to test such a value, an insn to do so
4245: can be constructed using this pattern.
1.1 root 4246:
1.1.1.2 root 4247: The output control string is a piece of C code which chooses which
4248: output template to return based on the kind of operand and the specific
4249: type of CPU for which code is being generated.
1.1 root 4250:
1.1.1.2 root 4251: @samp{"rm"} is an operand constraint. Its meaning is explained below.
1.1 root 4252:
1.1.1.2 root 4253: @node RTL Template, Output Template, Example, Machine Desc
4254: @section RTL Template for Generating and Recognizing Insns
1.1 root 4255:
1.1.1.2 root 4256: The RTL template is used to define which insns match the particular pattern
4257: and how to find their operands. For named patterns, the RTL template also
4258: says how to construct an insn from specified operands.
4259:
4260: Construction involves substituting specified operands into a copy of the
4261: template. Matching involves determining the values that serve as the
4262: operands in the insn being matched. Both of these activities are
4263: controlled by special expression types that direct matching and
4264: substitution of the operands.
1.1 root 4265:
4266: @table @code
4267: @item (match_operand:@var{m} @var{n} @var{testfn} @var{constraint})
4268: This expression is a placeholder for operand number @var{n} of
4269: the insn. When constructing an insn, operand number @var{n}
4270: will be substituted at this point. When matching an insn, whatever
4271: appears at this position in the insn will be taken as operand
4272: number @var{n}; but it must satisfy @var{testfn} or this instruction
4273: pattern will not match at all.
4274:
4275: Operand numbers must be chosen consecutively counting from zero in
4276: each instruction pattern. There may be only one @samp{match_operand}
4277: expression in the pattern for each expression number, and they must
4278: appear in order of increasing expression number.
4279:
4280: @var{testfn} is a string that is the name of a C function that accepts
4281: two arguments, a machine mode and an expression. During matching,
4282: the function will be called with @var{m} as the mode argument
4283: and the putative operand as the other argument. If it returns zero,
4284: this instruction pattern fails to match. @var{testfn} may be
4285: an empty string; then it means no test is to be done on the operand.
4286:
4287: Most often, @var{testfn} is @code{"general_operand"}. It checks
4288: that the putative operand is either a constant, a register or a
4289: memory reference, and that it is valid for mode @var{m}.
4290:
1.1.1.2 root 4291: For an operand that must be a register, @var{testfn} should be
4292: @code{"register_operand"}. This prevents GNU CC from creating insns
4293: that have memory references in these operands, insns which would only
4294: have to be taken apart in the reload pass.
4295:
4296: For an operand that must be a constant, either @var{testfn} should be
4297: @code{"immediate_operand"}, or the instruction pattern's extra condition
4298: should check for constants, or both.
4299:
4300: @var{constraint} is explained later (@pxref{Constraints}).
1.1 root 4301:
4302: @item (match_dup @var{n})
4303: This expression is also a placeholder for operand number @var{n}.
4304: It is used when the operand needs to appear more than once in the
4305: insn.
4306:
4307: In construction, @samp{match_dup} behaves exactly like
1.1.1.2 root 4308: @samp{match_operand}: the operand is substituted into the insn being
1.1 root 4309: constructed. But in matching, @samp{match_dup} behaves differently.
4310: It assumes that operand number @var{n} has already been determined by
1.1.1.2 root 4311: a @samp{match_operand} appearing earlier in the recognition template,
1.1 root 4312: and it matches only an identical-looking expression.
4313:
4314: @item (address (match_operand:@var{m} @var{n} "address_operand" ""))
4315: This complex of expressions is a placeholder for an operand number
4316: @var{n} in a ``load address'' instruction: an operand which specifies
4317: a memory location in the usual way, but for which the actual operand
4318: value used is the address of the location, not the contents of the
4319: location.
4320:
4321: @samp{address} expressions never appear in RTL code, only in machine
4322: descriptions. And they are used only in machine descriptions that do
4323: not use the operand constraint feature. When operand constraints are
4324: in use, the letter @samp{p} in the constraint serves this purpose.
4325:
4326: @var{m} is the machine mode of the @emph{memory location being
4327: addressed}, not the machine mode of the address itself. That mode is
4328: always the same on a given target machine (it is @code{Pmode}, which
4329: normally is @code{SImode}), so there is no point in mentioning it;
4330: thus, no machine mode is written in the @samp{address} expression. If
4331: some day support is added for machines in which addresses of different
4332: kinds of objects appear differently or are used differently (such as
4333: the PDP-10), different formats would perhaps need different machine
4334: modes and these modes might be written in the @samp{address}
4335: expression.
4336: @end table
4337:
1.1.1.2 root 4338: @node Output Template, Output Statement, RTL Template, Machine Desc
4339: @section Output Templates and Operand Substitution
1.1 root 4340:
1.1.1.2 root 4341: The @dfn{output template} is a string which specifies how to output
4342: the assembler code for an instruction pattern. Most of the template
4343: is a fixed string which is output literally. The character @samp{%}
4344: is used to specify where to substitute an operand; it can also be
4345: used to identify places different variants of the assembler require
4346: different syntax.
4347:
4348: In the simplest case, a @samp{%} followed by a digit @var{n} says to output
4349: operand @var{n} at that point in the string.
4350:
4351: @samp{%} followed by a letter and a digit says to output an operand in an
4352: alternate fashion. Four letters have standard, built-in meanings described
4353: below. The machine description macro @code{PRINT_OPERAND} can define
4354: additional letters with nonstandard meanings.
4355:
4356: @samp{%c@var{digit}} can be used to substitute an operand that is a
4357: constant value without the syntax that normally indicates an immediate
4358: operand.
4359:
4360: @samp{%n@var{digit}} is like @samp{%c@var{digit}} except that the value of
4361: the constant is negated before printing.
4362:
4363: @samp{%a@var{digit}} can be used to substitute an operand as if it were a
4364: memory reference, with the actual operand treated as the address. This may
4365: be useful when outputting a ``load address'' instruction, because often the
4366: assembler syntax for such an instruction requires you to write the operand
4367: as if it were a memory reference.
4368:
4369: @samp{%l@var{digit}} is used to substitute a @code{label_ref} into a jump
4370: instruction.
4371:
4372: @samp{%} followed by a punctuation character specifies a substitution that
4373: does not use an operand. Only one case is standard: @samp{%%} outputs a
4374: @samp{%} into the assembler code. Other nonstandard cases can be
4375: defined in the @code{PRINT_OPERAND} macro.
4376:
4377: The template may generate multiple assembler instructions. Write the text
4378: for the instructions, with @samp{\;} between them.
4379:
4380: When the RTL contains two operand which are required by constraint to match
4381: each other, the output template must refer only to the lower-numbered operand.
4382: Matching operands are not always identical, and the rest of the compiler
4383: arranges to put the proper RTL expression for printing into the lower-numbered
4384: operand.
4385:
4386: One use of nonstandard letters or punctuation following @samp{%} is to
4387: distinguish between different assembler languages for the same machine; for
4388: example, Motorola syntax versus MIT syntax for the 68000. Motorola syntax
4389: requires periods in most opcode names, while MIT syntax does not. For
4390: example, the opcode @samp{movel} in MIT syntax is @samp{move.l} in Motorola
4391: syntax. The same file of patterns is used for both kinds of output syntax,
4392: but the character sequence @samp{%.} is used in each place where Motorola
4393: syntax wants a period. The @code{PRINT_OPERAND} macro for Motorola syntax
4394: defines the sequence to output a period; the macro for MIT syntax defines
4395: it to do nothing.
4396:
4397: @node Output Statement, Constraints, Output Template, Machine Desc
4398: @section C Statements for Generating Assembler Output
4399:
4400: Often a single fixed template string cannot produce correct and efficient
4401: assembler code for all the cases that are recognized by a single
4402: instruction pattern. For example, the opcodes may depend on the kinds of
4403: operands; or some unfortunate combinations of operands may require extra
4404: machine instructions.
4405:
4406: If the output control string starts with a @samp{*}, then it is not an
4407: output template but rather a piece of C program that should compute a
4408: template. It should execute a @code{return} statement to return the
4409: template-string you want. Most such templates use C string literals, which
4410: require doublequote characters to delimit them. To include these
4411: doublequote characters in the string, prefix each one with @samp{\}.
4412:
4413: The operands may be found in the array @code{operands}, whose C data type
4414: is @code{rtx []}.
4415:
4416: It is possible to output an assembler instruction and then go on to output
4417: or compute more of them, using the subroutine @code{output_asm_insn}. This
4418: receives two arguments: a template-string and a vector of operands. The
4419: vector may be @code{operands}, or it may be another array of @code{rtx}
4420: that you declare locally and initialize yourself.
4421:
4422: When an insn pattern has multiple alternatives in its constraints, often
4423: the appearance of the assembler code determined mostly by which alternative
4424: was matched. When this is so, the C code can test the variable
4425: @code{which_alternative}, which is the ordinal number of the alternative
4426: that was actually satisfied (0 for the first, 1 for the second alternative,
4427: etc.).
4428:
4429: For example, suppose there are two opcodes for storing zero, @samp{clrreg}
4430: for registers and @samp{clrmem} for memory locations. Here is how
4431: a pattern could use @code{which_alternative} to choose between them:
1.1 root 4432:
4433: @example
1.1.1.2 root 4434: (define_insn ""
4435: [(set (match_operand:SI 0 "general_operand" "r,m")
4436: (const_int 0))]
1.1 root 4437: ""
4438: "*
1.1.1.2 root 4439: return (which_alternative == 0
4440: ? \"clrreg %0\" : \"clrmem %0\");
4441: ")
1.1 root 4442: @end example
4443:
1.1.1.2 root 4444: @node Constraints, Standard Names, Output Statement, Machine Desc
1.1 root 4445: @section Operand Constraints
4446:
4447: Each @samp{match_operand} in an instruction pattern can specify a
4448: constraint for the type of operands allowed. Constraints can say whether
4449: an operand may be in a register, and which kinds of register; whether the
4450: operand can be a memory reference, and which kinds of address; whether the
4451: operand may be an immediate constant, and which possible values it may
4452: have. Constraints can also require two operands to match.
4453:
4454: @menu
4455: * Simple Constraints:: Basic use of constraints.
1.1.1.2 root 4456: * Multi-Alternative:: When an insn has two alternative constraint-patterns.
1.1 root 4457: * Class Preferences:: Constraints guide which hard register to put things in.
4458: * Modifiers:: More precise control over effects of constraints.
4459: * No Constraints:: Describing a clean machine without constraints.
4460: @end menu
4461:
4462: @node Simple Constraints, Multi-Alternative, Constraints, Constraints
4463: @subsection Simple Constraints
4464:
4465: The simplest kind of constraint is a string full of letters, each of
4466: which describes one kind of operand that is permitted. Here are
4467: the letters that are allowed:
4468:
1.1.1.2 root 4469: @table @asis
4470: @item @samp{m}
1.1 root 4471: A memory operand is allowed, with any kind of address that the machine
4472: supports in general.
4473:
1.1.1.2 root 4474: @item @samp{o}
4475: A memory operand is allowed, but only if the address is
4476: @dfn{offsetable}. This means that adding a small integer (actually,
4477: the width in bytes of the operand, as determined by its machine mode)
4478: may be added to the address and the result is also a valid memory
4479: address.
4480:
4481: For example, an address which is constant is offsetable; so is an
4482: address that is the sum of a register and a constant (as long as a
4483: slightly larger constant is also within the range of address-offsets
4484: supported by the machine); but an autoincrement or autodecrement
4485: address is not offsetable. More complicated indirect/indexed
4486: addresses may or may not be offsetable depending on the other
4487: addressing modes that the machine supports.
4488:
4489: Note that in an output operand which can be matched by another
4490: operand, the constraint letter @samp{o} is valid only when accompanied
4491: by both @samp{<} (if the target machine has predecrement addressing)
4492: and @samp{>} (if the target machine has preincrement addressing).
1.1 root 4493:
1.1.1.2 root 4494: @item @samp{<}
1.1 root 4495: A memory operand with autodecrement addressing (either predecrement or
4496: postdecrement) is allowed.
4497:
1.1.1.2 root 4498: @item @samp{>}
1.1 root 4499: A memory operand with autoincrement addressing (either preincrement or
4500: postincrement) is allowed.
4501:
1.1.1.2 root 4502: @item @samp{r}
4503: A register operand is allowed provided that it is in a general
4504: register.
1.1 root 4505:
1.1.1.2 root 4506: @item @samp{d}, @samp{a}, @samp{f}, @dots{}
1.1 root 4507: Other letters can be defined in machine-dependent fashion to stand for
4508: particular classes of registers. @samp{d}, @samp{a} and @samp{f} are
1.1.1.2 root 4509: defined on the 68000/68020 to stand for data, address and floating
4510: point registers.
1.1 root 4511:
1.1.1.2 root 4512: @item @samp{i}
1.1 root 4513: An immediate integer operand (one with constant value) is allowed.
1.1.1.2 root 4514: This includes symbolic constants whose values will be known only at
4515: assembly time.
1.1 root 4516:
1.1.1.2 root 4517: @item @samp{n}
4518: An immediate integer operand with a known numeric value is allowed.
4519: Many systems cannot support assembly-time constants for operands less
4520: than a word wide. Constraints for these operands should use @samp{n}
4521: rather than @samp{i}.
4522:
4523: @item @samp{I}, @samp{J}, @samp{K}, @dots{}
4524: Other letters in the range @samp{I} through @samp{M} may be defined in
4525: a machine-dependent fashion to permit immediate integer operands with
4526: explicit integer values in specified ranges. For example, on the
4527: 68000, @samp{I} is defined to stand for the range of values 1 to 8.
4528: This is the range permitted as a shift count in the shift
4529: instructions.
1.1 root 4530:
1.1.1.2 root 4531: @item @samp{F}
1.1 root 4532: An immediate floating operand (expression code @samp{const_double}) is
4533: allowed.
4534:
1.1.1.2 root 4535: @item @samp{G}, @samp{H}
1.1 root 4536: @samp{G} and @samp{H} may be defined in a machine-dependent fashion to
4537: permit immediate floating operands in particular ranges of values.
4538:
1.1.1.2 root 4539: @item @samp{s}
1.1 root 4540: An immediate integer operand whose value is not an explicit integer is
1.1.1.2 root 4541: allowed.
4542:
4543: This might appear strange; if an insn allows a constant operand with a
4544: value not known at compile time, it certainly must allow any known
1.1 root 4545: value. So why use @samp{s} instead of @samp{i}? Sometimes it allows
1.1.1.2 root 4546: better code to be generated.
4547:
4548: For example, on the 68000 in a fullword instruction it is possible to
4549: use an immediate operand; but if the immediate value is between -32
4550: and 31, better code results from loading the value into a register and
4551: using the register. This is because the load into the register can be
4552: done with a @samp{moveq} instruction. We arrange for this to happen
4553: by defining the letter @samp{K} to mean ``any integer outside the
4554: range -32 to 31'', and then specifying @samp{Ks} in the operand
1.1 root 4555: constraints.
4556:
1.1.1.2 root 4557: @item @samp{g}
1.1 root 4558: Any register, memory or immediate integer operand is allowed, except for
4559: registers that are not general registers.
4560:
1.1.1.2 root 4561: @item @samp{@var{n}} (a digit)
4562: An operand that matches operand number @var{n} is allowed.
1.1 root 4563: If a digit is used together with letters, the digit should come last.
4564:
1.1.1.2 root 4565: This is called a @dfn{matching constraint} and what it really means is
4566: that the assembler has only a single operand that fills two roles
4567: considered separate in the RTL insn. For example, an add insn has two
4568: input operands and one output operand in the RTL, but on most machines
4569: an add instruction really has only two operands, one of them an
4570: input-output operand.
4571:
4572: Matching constraints work only in circumstances like that add insn.
4573: More precisely, the matching constraint must appear in an input-only
4574: operand and the operand that it matches must be an output-only operand
4575: with a lower number.
4576:
4577: For operands to match in a particular case usually means that they
4578: are identical-looking RTL expressions. But in a few special cases
4579: specific kinds of dissimilarity are allowed. For example, @code{*x}
4580: as an input operand will match @code{*x++} as an output operand.
4581: For proper results in such cases, the output template should always
4582: use the output-operand's number when printing the operand.
4583:
4584: @item @samp{p}
1.1 root 4585: An operand that is a valid memory address is allowed. This is
4586: for ``load address'' and ``push address'' instructions.
4587:
4588: If @samp{p} is used in the constraint, the test-function in the
4589: @samp{match_operand} must be @code{address_operand}.
4590: @end table
4591:
4592: In order to have valid assembler code, each operand must satisfy
4593: its constraint. But a failure to do so does not prevent the pattern
4594: from applying to an insn. Instead, it directs the compiler to modify
1.1.1.2 root 4595: the code so that the constraint will be satisfied. Usually this is
1.1 root 4596: done by copying an operand into a register.
4597:
4598: Contrast, therefore, the two instruction patterns that follow:
4599:
4600: @example
4601: (define_insn ""
4602: [(set (match_operand:SI 0 "general_operand" "r")
4603: (plus:SI (match_dup 0)
4604: (match_operand:SI 1 "general_operand" "r")))]
4605: ""
4606: "@dots{}")
4607: @end example
4608:
4609: @noindent
4610: which has two operands, one of which must appear in two places, and
4611:
4612: @example
4613: (define_insn ""
4614: [(set (match_operand:SI 0 "general_operand" "r")
4615: (plus:SI (match_operand:SI 1 "general_operand" "0")
4616: (match_operand:SI 2 "general_operand" "r")))]
4617: ""
4618: "@dots{}")
4619: @end example
4620:
4621: @noindent
4622: which has three operands, two of which are required by a constraint to be
4623: identical. If we are considering an insn of the form
4624:
4625: @example
4626: (insn @var{n} @var{prev} @var{next}
4627: (set (reg:SI 3)
4628: (plus:SI (reg:SI 6) (reg:SI 109)))
4629: @dots{})
4630: @end example
4631:
4632: @noindent
4633: the first pattern would not apply at all, because this insn does not
4634: contain two identical subexpressions in the right place. The pattern would
4635: say, ``That does not look like an add instruction; try other patterns.''
4636: The second pattern would say, ``Yes, that's an add instruction, but there
4637: is something wrong with it.'' It would direct the reload pass of the
4638: compiler to generate additional insns to make the constraint true. The
4639: results might look like this:
4640:
4641: @example
4642: (insn @var{n2} @var{prev} @var{n}
4643: (set (reg:SI 3) (reg:SI 6))
4644: @dots{})
4645:
4646: (insn @var{n} @var{n2} @var{next}
4647: (set (reg:SI 3)
4648: (plus:SI (reg:SI 3) (reg:SI 109)))
4649: @dots{})
4650: @end example
4651:
4652: Because insns that don't fit the constraints are fixed up by loading
4653: operands into registers, every instruction pattern's constraints must
4654: permit the case where all the operands are in registers. It need not
4655: permit all classes of registers; the compiler knows how to copy registers
4656: into other registers of the proper class in order to make an instruction
4657: valid. But if no registers are permitted, the compiler will be stymied: it
4658: does not know how to save a register in memory in order to make an
4659: instruction valid. Instruction patterns that reject registers can be
4660: made valid by attaching a condition-expression that refuses to match
4661: an insn at all if the crucial operand is a register.
4662:
4663: @node Multi-Alternative, Class Preferences, Simple Constraints, Constraints
4664: @subsection Multiple Alternative Constraints
4665:
4666: Sometimes a single instruction has multiple alternative sets of possible
4667: operands. For example, on the 68000, a logical-or instruction can combine
4668: register or an immediate value into memory, or it can combine any kind of
4669: operand into a register; but it cannot combine one memory location into
4670: another.
4671:
4672: These constraints are represented as multiple alternatives. An alternative
4673: can be described by a series of letters for each operand. The overall
4674: constraint for an operand is made from the letters for this operand
4675: from the first alternative, a comma, the letters for this operand from
4676: the second alternative, a comma, and so on until the last alternative.
4677: Here is how it is done for fullword logical-or on the 68000:
4678:
4679: @example
4680: (define_insn "iorsi3"
4681: [(set (match_operand:SI 0 "general_operand" "=%m,d")
1.1.1.2 root 4682: (ior:SI (match_operand:SI 1 "general_operand" "0,0")
4683: (match_operand:SI 2 "general_operand" "dKs,dmKs")))]
1.1 root 4684: @dots{})
4685: @end example
4686:
4687: The first alternative has @samp{m} (memory) for operand 0, @samp{0} for
4688: operand 1 (meaning it must match operand 0), and @samp{dKs} for operand 2.
4689: The second alternative has @samp{d} (data register) for operand 0, @samp{0}
4690: for operand 1, and @samp{dmKs} for operand 2. The @samp{=} and @samp{%} in
4691: the constraint for operand 0 are not part of any alternative; their meaning
4692: is explained in the next section.
4693:
4694: If all the operands fit any one alternative, the instruction is valid.
4695: Otherwise, for each alternative, the compiler counts how many instructions
4696: must be added to copy the operands so that that alternative applies.
4697: The alternative requiring the least copying is chosen. If two alternatives
4698: need the same amount of copying, the one that comes first is chosen.
4699: These choices can be altered with the @samp{?} and @samp{!} characters:
4700:
4701: @table @samp
4702: @item ?
4703: Disparage slightly the alternative that the @samp{?} appears in,
4704: as a choice when no alternative applies exactly. The compiler regards
4705: this alternative as one unit more costly for each @samp{?} that appears
4706: in it.
4707:
4708: @item !
4709: Disparage severely the alternative that the @samp{!} appears in.
4710: When operands must be copied into registers, the compiler will
4711: never choose this alternative as the one to strive for.
4712: @end table
4713:
1.1.1.2 root 4714: When an insn pattern has multiple alternatives in its constraints,
4715: often the appearance of the assembler code determined mostly by which
4716: alternative was matched. When this is so, the C code for writing the
4717: assembler code can use the variable @code{which_alternative}, which is
4718: the ordinal number of the alternative that was actually satisfied
4719: (0 for the first, 1 for the second alternative, etc.). For example:
4720:
4721: @example
4722: (define_insn ""
4723: [(set (match_operand:SI 0 "general_operand" "r,m")
4724: (const_int 0))]
4725: ""
4726: "*
4727: return (which_alternative == 0
4728: ? \"clrreg %0\" : \"clrmem %0\");
4729: ")
4730: @end example
4731:
1.1 root 4732: @node Class Preferences, Modifiers, Multi-Alternative, Constraints
4733: @subsection Register Class Preferences
4734:
4735: The operand constraints have another function: they enable the compiler
4736: to decide which kind of hardware register a pseudo register is best
4737: allocated to. The compiler examines the constraints that apply to the
4738: insns that use the pseudo register, looking for the machine-dependent
4739: letters such as @samp{d} and @samp{a} that specify classes of registers.
4740: The pseudo register is put in whichever class gets the most ``votes''.
4741: The constraint letters @samp{g} and @samp{r} also vote: they vote in
4742: favor of a general register. The machine description says which registers
4743: are considered general.
4744:
4745: Of course, on some machines all registers are equivalent, and no register
4746: classes are defined. Then none of this complexity is relevant.
4747:
4748: @node Modifiers, No Constraints, Class Preferences, Constraints
4749: @subsection Constraint Modifier Characters
4750:
4751: @table @samp
4752: @item =
1.1.1.2 root 4753: Means that this operand is write-only for this instruction: the previous
4754: value is discarded and replaced by output data.
1.1 root 4755:
4756: @item +
4757: Means that this operand is both read and written by the instruction.
4758:
4759: When the compiler fixes up the operands to satisfy the constraints,
4760: it needs to know which operands are inputs to the instruction and
4761: which are outputs from it. @samp{=} identifies an output; @samp{+}
4762: identifies an operand that is both input and output; all other operands
4763: are assumed to be input only.
4764:
1.1.1.2 root 4765: @item &
4766: Means (in a particular alternative) that this operand is written
4767: before the instruction is finished using the input operands.
4768: Therefore, this operand may not lie in a register that is used as an
4769: input operand or as part of any memory address.
4770:
4771: @samp{&} applies only to the alternative in which it is written. In
4772: constraints with multiple alternatives, sometimes one alternative
4773: requires @samp{&} while others do not. See, for example, the
4774: @samp{movdf} insn of the 68000.
4775:
4776: @samp{&} does not obviate the need to write @samp{=}.
4777:
1.1 root 4778: @item %
1.1.1.2 root 4779: Declares the instruction to be commutative for this operand and the
4780: following operand. This means that the compiler may interchange the
4781: two operands if that is the cheapest way to make all operands fit the
4782: constraints. This is often used in patterns for addition instructions
4783: that really have only two operands: the result must go in one of the
4784: arguments. Here for example, is how the 68000 halfword-add
4785: instruction is defined:
4786:
4787: @example
4788: (define_insn "addhi3"
4789: [(set (match_operand:HI 0 "general_operand" "=m,r")
4790: (plus:HI (match_operand:HI 1 "general_operand" "%0,0")
4791: (match_operand:HI 2 "general_operand" "di,g")))]
4792: @dots{})
4793: @end example
4794:
4795: Note that in previous versions of GNU CC the @samp{%} constraint
4796: modifier always applied to operands 1 and 2 regardless of which
4797: operand it was written in. The usual custom was to write it in
4798: operand 0. Now it must be in operand 1 if the operands to be
4799: exchanged are 1 and 2.
1.1 root 4800:
4801: @item #
1.1.1.2 root 4802: Says that all following characters, up to the next comma, are to be
4803: ignored as a constraint. They are significant only for choosing
4804: register preferences.
1.1 root 4805:
4806: @item *
4807: Says that the following character should be ignored when choosing
1.1.1.2 root 4808: register preferences. @samp{*} has no effect on the meaning of the
4809: constraint as a constraint.
4810:
4811: Here is an example: the 68000 has an instruction to sign-extend a
4812: halfword in a data register, and can also sign-extend a value by
4813: copying it into an address register. While either kind of register is
4814: acceptable, the constraints on an address-register destination are
4815: less strict, so it is best if register allocation makes an address
4816: register its goal. Therefore, @samp{*} is used so that the @samp{d}
4817: constraint letter (for data register) is ignored when computing
4818: register preferences.
4819:
4820: @example
4821: (define_insn "extendhisi2"
4822: [(set (match_operand:SI 0 "general_operand" "=*d,a")
4823: (sign_extend:SI
4824: (match_operand:HI 1 "general_operand" "0,g")))]
4825: @dots{})
4826: @end example
1.1 root 4827: @end table
4828:
4829: @node No Constraints,, Modifiers, Constraints
4830: @subsection Not Using Constraints
4831:
4832: Some machines are so clean that operand constraints are not required. For
4833: example, on the Vax, an operand valid in one context is valid in any other
1.1.1.2 root 4834: context. On such a machine, every operand constraint would be @samp{g},
1.1 root 4835: excepting only operands of ``load address'' instructions which are
4836: written as if they referred to a memory location's contents but actual
1.1.1.2 root 4837: refer to its address. They would have constraint @samp{p}.
1.1 root 4838:
1.1.1.2 root 4839: For such machines, instead of writing @samp{g} and @samp{p} for all
1.1 root 4840: the constraints, you can choose to write a description with empty constraints.
4841: Then you write @samp{""} for the constraint in every @samp{match_operand}.
4842: Address operands are identified by writing an @samp{address} expression
4843: around the @samp{match_operand}, not by their constraints.
4844:
4845: When the machine description has just empty constraints, certain parts
4846: of compilation are skipped, making the compiler faster.
4847:
1.1.1.2 root 4848: @node Standard Names, Pattern Ordering, Constraints, Machine Desc
4849: @section Standard Names for Patterns Used in Generation
1.1 root 4850:
4851: Here is a table of the instruction names that are meaningful in the RTL
4852: generation pass of the compiler. Giving one of these names to an
4853: instruction pattern tells the RTL generation pass that it can use the
4854: pattern in to accomplish a certain task.
4855:
1.1.1.2 root 4856: @table @asis
4857: @item @samp{mov@var{m}}
1.1 root 4858: Here @var{m} is a two-letter machine mode name, in lower case. This
4859: instruction pattern moves data with that machine mode from operand 1 to
4860: operand 0. For example, @samp{movsi} moves full-word data.
4861:
4862: If operand 0 is a @samp{subreg} with mode @var{m} of a register whose
4863: natural mode is wider than @var{m}, the effect of this instruction is
4864: to store the specified value in the part of the register that corresponds
4865: to mode @var{m}. The effect on the rest of the register is undefined.
4866:
1.1.1.3 root 4867: This class of patterns is special in several ways. First of all, each
4868: of these names @emph{must} be defined, because there is no other way
4869: to copy a datum from one place to another.
4870:
4871: Second, these patterns are not used solely in the RTL generation pass.
4872: Even the reload pass can generate move insns to copy values from stack
4873: slots into temporary registers. When it does so, one of the operands
4874: is a hard register and the other is an operand that can have a reload.
4875:
4876: Therefore, when given such a pair of operands, the pattern must
4877: generate RTL which needs no temporary registers---no registers other
4878: than the operands. For example, if you support the pattern with a
4879: @code{define_expand}, then in such a case you mustn't call
4880: @code{force_reg} or any other such function which might generate new
4881: pseudo registers.
4882:
4883: This requirement exists even for subword modes on a RISC machine where
4884: fetching those modes from memory normally requires several insns and
4885: some temporary registers. Look in @file{spur.md} to see how the
4886: requirement is satisfied.
4887:
4888: The variety of operands that have reloads depends on the rest of the
4889: machine description, but typically on a RISC machine these can only be
4890: pseudo registers that did not get hard registers, while on other
4891: machines explicit memory references will get optional reloads.
4892:
1.1.1.2 root 4893: @item @samp{movstrict@var{m}}
1.1 root 4894: Like @samp{mov@var{m}} except that if operand 0 is a @samp{subreg}
4895: with mode @var{m} of a register whose natural mode is wider,
4896: the @samp{movstrict@var{m}} instruction is guaranteed not to alter
4897: any of the register except the part which belongs to mode @var{m}.
4898:
1.1.1.2 root 4899: @item @samp{add@var{m}3}
1.1 root 4900: Add operand 2 and operand 1, storing the result in operand 0. All operands
4901: must have mode @var{m}. This can be used even on two-address machines, by
4902: means of constraints requiring operands 1 and 0 to be the same location.
4903:
1.1.1.2 root 4904: @item @samp{sub@var{m}3}, @samp{mul@var{m}3}, @samp{umul@var{m}3}, @samp{div@var{m}3}, @samp{udiv@var{m}3}, @samp{mod@var{m}3}, @samp{umod@var{m}3}, @samp{and@var{m}3}, @samp{ior@var{m}3}, @samp{xor@var{m}3}
1.1 root 4905: Similar, for other arithmetic operations.
4906:
1.1.1.2 root 4907: @item @samp{andcb@var{m}3}
1.1 root 4908: Bitwise logical-and operand 1 with the complement of operand 2
4909: and store the result in operand 0.
4910:
1.1.1.2 root 4911: @item @samp{mulhisi3}
1.1 root 4912: Multiply operands 1 and 2, which have mode @code{HImode}, and store
4913: a @code{SImode} product in operand 0.
4914:
1.1.1.2 root 4915: @item @samp{mulqihi3}, @samp{mulsidi3}
1.1 root 4916: Similar widening-multiplication instructions of other widths.
4917:
1.1.1.2 root 4918: @item @samp{umulqihi3}, @samp{umulhisi3}, @samp{umulsidi3}
1.1 root 4919: Similar widening-multiplication instructions that do unsigned
4920: multiplication.
4921:
1.1.1.2 root 4922: @item @samp{divmod@var{m}4}
1.1 root 4923: Signed division that produces both a quotient and a remainder.
4924: Operand 1 is divided by operand 2 to produce a quotient stored
4925: in operand 0 and a remainder stored in operand 3.
4926:
1.1.1.2 root 4927: @item @samp{udivmod@var{m}4}
1.1 root 4928: Similar, but does unsigned division.
4929:
1.1.1.2 root 4930: @item @samp{divmod@var{m}@var{n}4}
1.1 root 4931: Like @samp{divmod@var{m}4} except that only the dividend has mode
4932: @var{m}; the divisor, quotient and remainder have mode @var{n}.
4933: For example, the Vax has a @samp{divmoddisi4} instruction
4934: (but it is omitted from the machine description, because it
4935: is so slow that it is faster to compute remainders by the
4936: circumlocution that the compiler will use if this instruction is
4937: not available).
4938:
1.1.1.2 root 4939: @item @samp{ashl@var{m}3}
1.1 root 4940: Arithmetic-shift operand 1 left by a number of bits specified by
4941: operand 2, and store the result in operand 0. Operand 2 has
4942: mode @code{SImode}, not mode @var{m}.
4943:
1.1.1.2 root 4944: @item @samp{ashr@var{m}3}, @samp{lshl@var{m}3}, @samp{lshr@var{m}3}, @samp{rotl@var{m}3}, @samp{rotr@var{m}3}
1.1 root 4945: Other shift and rotate instructions.
4946:
1.1.1.2 root 4947: Logical and arithmetic left shift are the same. Machines that do not
4948: allow negative shift counts often have only one instruction for
4949: shifting left. On such machines, you should define a pattern named
4950: @samp{ashl@var{m}3} and leave @samp{lshl@var{m}3} undefined.
4951:
4952: @item @samp{neg@var{m}2}
1.1 root 4953: Negate operand 1 and store the result in operand 0.
4954:
1.1.1.2 root 4955: @item @samp{abs@var{m}2}
1.1 root 4956: Store the absolute value of operand 1 into operand 0.
4957:
1.1.1.2 root 4958: @item @samp{sqrt@var{m}2}
1.1 root 4959: Store the square root of operand 1 into operand 0.
4960:
1.1.1.2 root 4961: @item @samp{ffs@var{m}2}
4962: Store into operand 0 one plus the index of the least significant 1-bit
4963: of operand 1. If operand 1 is zero, store zero. @var{m} is the mode
4964: of operand 0; operand 1's mode is specified by the instruction
4965: pattern, and the compiler will convert the operand to that mode before
4966: generating the instruction.
4967:
4968: @item @samp{one_cmpl@var{m}2}
1.1 root 4969: Store the bitwise-complement of operand 1 into operand 0.
4970:
1.1.1.2 root 4971: @item @samp{cmp@var{m}}
1.1 root 4972: Compare operand 0 and operand 1, and set the condition codes.
1.1.1.2 root 4973: The RTL pattern should look like this:
1.1 root 4974:
1.1.1.2 root 4975: @example
4976: (set (cc0) (minus (match_operand:@var{m} 0 @dots{})
4977: (match_operand:@var{m} 1 @dots{})))
4978: @end example
4979:
4980: Each such definition in the machine description, for integer mode
4981: @var{m}, must have a corresponding @samp{tst@var{m}} pattern, because
4982: optimization can simplify the compare into a test when operand 1 is
4983: zero.
4984:
4985: @item @samp{tst@var{m}}
1.1 root 4986: Compare operand 0 against zero, and set the condition codes.
1.1.1.2 root 4987: The RTL pattern should look like this:
1.1 root 4988:
1.1.1.2 root 4989: @example
4990: (set (cc0) (match_operand:@var{m} 0 @dots{}))
4991: @end example
4992:
4993: @item @samp{movstr@var{m}}
1.1 root 4994: Block move instruction. The addresses of the destination and source
4995: strings are the first two operands, and both are in mode @code{Pmode}.
4996: The number of bytes to move is the third operand, in mode @var{m}.
4997:
1.1.1.2 root 4998: @item @samp{cmpstr@var{m}}
1.1 root 4999: Block compare instruction, with operands like @samp{movstr@var{m}}
5000: except that the two memory blocks are compared byte by byte
5001: in lexicographic order. The effect of the instruction is to set
5002: the condition codes.
5003:
1.1.1.2 root 5004: @item @samp{float@var{m}@var{n}2}
1.1 root 5005: Convert operand 1 (valid for fixed point mode @var{m}) to floating
5006: point mode @var{n} and store in operand 0 (which has mode @var{n}).
5007:
1.1.1.2 root 5008: @item @samp{fix@var{m}@var{n}2}
5009: Convert operand 1 (valid for floating point mode @var{m}) to fixed
5010: point mode @var{n} as a signed number and store in operand 0 (which
5011: has mode @var{n}). This instruction's result is defined only when
5012: the value of operand 1 is an integer.
5013:
5014: @item @samp{fixuns@var{m}@var{n}2}
5015: Convert operand 1 (valid for floating point mode @var{m}) to fixed
5016: point mode @var{n} as an unsigned number and store in operand 0 (which
5017: has mode @var{n}). This instruction's result is defined only when the
5018: value of operand 1 is an integer.
5019:
5020: @item @samp{ftrunc@var{m}2}
5021: Convert operand 1 (valid for floating point mode @var{m}) to an
5022: integer value, still represented in floating point mode @var{m}, and
5023: store it in operand 0 (valid for floating point mode @var{m}).
5024:
5025: @item @samp{fix_trunc@var{m}@var{n}2}
5026: Like @samp{fix@var{m}@var{n}2} but works for any floating point value
5027: of mode @var{m} by converting the value to an integer.
5028:
5029: @item @samp{fixuns_trunc@var{m}@var{n}2}
5030: Like @samp{fixuns@var{m}@var{n}2} but works for any floating point
5031: value of mode @var{m} by converting the value to an integer.
5032:
5033: @item @samp{trunc@var{m}@var{n}}
1.1 root 5034: Truncate operand 1 (valid for mode @var{m}) to mode @var{n} and
5035: store in operand 0 (which has mode @var{n}). Both modes must be fixed
5036: point or both floating point.
5037:
1.1.1.2 root 5038: @item @samp{extend@var{m}@var{n}}
1.1 root 5039: Sign-extend operand 1 (valid for mode @var{m}) to mode @var{n} and
5040: store in operand 0 (which has mode @var{n}). Both modes must be fixed
5041: point or both floating point.
5042:
1.1.1.2 root 5043: @item @samp{zero_extend@var{m}@var{n}}
1.1 root 5044: Zero-extend operand 1 (valid for mode @var{m}) to mode @var{n} and
5045: store in operand 0 (which has mode @var{n}). Both modes must be fixed
5046: point.
5047:
1.1.1.2 root 5048: @item @samp{extv}
1.1 root 5049: Extract a bit-field from operand 1 (a register or memory operand),
5050: where operand 2 specifies the width in bits and operand 3 the starting
5051: bit, and store it in operand 0. Operand 0 must have @code{Simode}.
5052: Operand 1 may have mode @code{QImode} or @code{SImode}; often
5053: @code{SImode} is allowed only for registers. Operands 2 and 3 must be
5054: valid for @code{SImode}.
5055:
5056: The RTL generation pass generates this instruction only with constants
5057: for operands 2 and 3.
5058:
5059: The bit-field value is sign-extended to a full word integer
5060: before it is stored in operand 0.
5061:
1.1.1.2 root 5062: @item @samp{extzv}
1.1 root 5063: Like @samp{extv} except that the bit-field value is zero-extended.
5064:
1.1.1.2 root 5065: @item @samp{insv}
1.1 root 5066: Store operand 3 (which must be valid for @code{SImode}) into a
5067: bit-field in operand 0, where operand 1 specifies the width in bits
5068: and operand 2 the starting bit. Operand 0 may have mode @code{QImode}
5069: or @code{SImode}; often @code{SImode} is allowed only for registers.
5070: Operands 1 and 2 must be valid for @code{SImode}.
5071:
5072: The RTL generation pass generates this instruction only with constants
5073: for operands 1 and 2.
5074:
1.1.1.2 root 5075: @item @samp{s@var{cond}}
5076: Store zero or nonzero in the operand according to the condition codes.
5077: Value stored is nonzero iff the condition @var{cond} is true.
5078: @var{cond} is the name of a comparison operation expression code, such
1.1 root 5079: as @samp{eq}, @samp{lt} or @samp{leu}.
5080:
1.1.1.2 root 5081: You specify the mode that the operand must have when you write the
5082: @code{match_operand} expression. The compiler automatically sees
5083: which mode you have used and supplies an operand of that mode.
5084:
5085: The value stored for a true condition must have 1 as its low bit.
5086: Otherwise the instruction is not suitable and must be omitted from the
5087: machine description. You must tell the compiler exactly which value
5088: is stored by defining the macro @code{STORE_FLAG_VALUE}.
5089:
5090: @item @samp{b@var{cond}}
1.1 root 5091: Conditional branch instruction. Operand 0 is a @samp{label_ref}
5092: that refers to the label to jump to. Jump if the condition codes
5093: meet condition @var{cond}.
5094:
1.1.1.2 root 5095: @item @samp{call}
5096: Subroutine call instruction. Operand 1 is the number of bytes of
5097: arguments pushed (in mode @code{SImode}), and operand 0 is the
5098: function to call. Operand 0 should be a @samp{mem} RTX whose address
5099: is the address of the function.
1.1 root 5100:
1.1.1.2 root 5101: @item @samp{return}
1.1 root 5102: Subroutine return instruction. This instruction pattern name should be
5103: defined only if a single instruction can do all the work of returning
5104: from a function.
5105:
1.1.1.3 root 5106: @item @samp{casesi}
5107: Instruction to jump through a dispatch table, including bounds checking.
5108: This instruction takes five operands:
5109:
5110: @enumerate
5111: @item
5112: The index to dispatch on, which has mode @code{SImode}.
5113:
5114: @item
5115: The lower bound for indices in the table, an integer constant.
5116:
5117: @item
5118: The upper bound for indices in the table, an integer constant.
5119:
5120: @item
5121: A label to jump to if the index has a value outside the bounds.
5122: (If the machine-description macro @code{CASE_DROPS_THROUGH} is defined,
5123: then an out-of-bounds index drops through to the code following
5124: the jump table instead of jumping to this label. In that case,
5125: this label is not actually used by the @samp{casesi} instruction,
5126: but it is always provided as an operand.)
5127:
5128: @item
5129: A label that precedes the table itself.
5130: @end enumerate
5131:
5132: The table is a @samp{addr_vec} or @samp{addr_diff_vec} inside of a
5133: @samp{jump_insn}. The number of elements in the table is one plus the
5134: difference between the upper bound and the lower bound.
5135:
1.1.1.2 root 5136: @item @samp{tablejump}
1.1.1.3 root 5137: Instruction to jump to a variable address. This is a low-level
5138: capability which can be used to implement a dispatch table when there
5139: is no @samp{casesi} pattern.
5140:
5141: This pattern requires two operands: the address or offset, and a label
5142: which should immediately precede the jump table. If the macro
5143: @code{CASE_VECTOR_PC_RELATIVE} is defined then the first operand is an
5144: absolute address to jump to; otherwise, it is an offset which counts
5145: from the address of the table.
5146:
5147: The @samp{tablejump} insn is always the last insn before the jump
5148: table it uses. Its assembler code normally has no need to use the
5149: second operand, but you should incorporate it in the RTL pattern so
5150: that the jump optimizer will not delete the table as unreachable code.
1.1 root 5151: @end table
5152:
1.1.1.2 root 5153: @node Pattern Ordering, Dependent Patterns, Standard Names, Machine Desc
5154: @section When the Order of Patterns Matters
5155:
5156: Sometimes an insn can match more than one instruction pattern. Then the
5157: pattern that appears first in the machine description is the one used.
5158: Therefore, more specific patterns (patterns that will match fewer things)
5159: and faster instructions (those that will produce better code when they
5160: do match) should usually go first in the description.
5161:
5162: In some cases the effect of ordering the patterns can be used to hide
5163: a pattern when it is not valid. For example, the 68000 has an
5164: instruction for converting a fullword to floating point and another
5165: for converting a byte to floating point. An instruction converting
5166: an integer to floating point could match either one. We put the
5167: pattern to convert the fullword first to make sure that one will
5168: be used rather than the other. (Otherwise a large integer might
5169: be generated as a single-byte immediate quantity, which would not work.)
5170: Instead of using this pattern ordering it would be possible to make the
5171: pattern for convert-a-byte smart enough to deal properly with any
5172: constant value.
5173:
5174: @node Dependent Patterns, Jump Patterns, Pattern Ordering, Machine Desc
5175: @section Interdependence of Patterns
1.1 root 5176:
5177: Every machine description must have a named pattern for each of the
5178: conditional branch names @samp{b@var{cond}}. The recognition template
5179: must always have the form
5180:
5181: @example
5182: (set (pc)
5183: (if_then_else (@var{cond} (cc0) (const_int 0))
5184: (label_ref (match_operand 0 "" ""))
5185: (pc)))
5186: @end example
5187:
5188: @noindent
5189: In addition, every machine description must have an anonymous pattern
5190: for each of the possible reverse-conditional branches. These patterns
5191: look like
5192:
5193: @example
5194: (set (pc)
5195: (if_then_else (@var{cond} (cc0) (const_int 0))
5196: (pc)
5197: (label_ref (match_operand 0 "" ""))))
5198: @end example
5199:
5200: @noindent
5201: They are necessary because jump optimization can turn direct-conditional
5202: branches into reverse-conditional branches.
5203:
5204: The compiler does more with RTL than just create it from patterns
5205: and recognize the patterns: it can perform arithmetic expression codes
5206: when constant values for their operands can be determined. As a result,
5207: sometimes having one pattern can require other patterns. For example, the
5208: Vax has no `and' instruction, but it has `and not' instructions. Here
5209: is the definition of one of them:
5210:
5211: @example
5212: (define_insn "andcbsi2"
5213: [(set (match_operand:SI 0 "general_operand" "")
5214: (and:SI (match_dup 0)
5215: (not:SI (match_operand:SI
5216: 1 "general_operand" ""))))]
5217: ""
5218: "bicl2 %1,%0")
5219: @end example
5220:
5221: @noindent
5222: If operand 1 is an explicit integer constant, an instruction constructed
1.1.1.2 root 5223: using that pattern can be simplified into an `and' like this:
1.1 root 5224:
5225: @example
5226: (set (reg:SI 41)
5227: (and:SI (reg:SI 41)
5228: (const_int 0xffff7fff)))
5229: @end example
5230:
5231: @noindent
5232: (where the integer constant is the one's complement of what
5233: appeared in the original instruction).
5234:
5235: To avoid a fatal error, the compiler must have a pattern that recognizes
5236: such an instruction. Here is what is used:
5237:
5238: @example
5239: (define_insn ""
5240: [(set (match_operand:SI 0 "general_operand" "")
5241: (and:SI (match_dup 0)
5242: (match_operand:SI 1 "general_operand" "")))]
5243: "GET_CODE (operands[1]) == CONST_INT"
5244: "*
1.1.1.2 root 5245: @{ operands[1]
1.1 root 5246: = gen_rtx (CONST_INT, VOIDmode, ~INTVAL (operands[1]));
5247: return \"bicl2 %1,%0\";
1.1.1.2 root 5248: @}")
1.1 root 5249: @end example
5250:
5251: @noindent
5252: Whereas a pattern to match a general `and' instruction is impossible to
5253: support on the Vax, this pattern is possible because it matches only a
5254: constant second argument: a special case that can be output as an `and not'
5255: instruction.
5256:
1.1.1.2 root 5257: A ``compare'' instruction whose RTL looks like this:
5258:
5259: @example
5260: (set (cc0) (minus @var{operand} (const_int 0)))
5261: @end example
5262:
5263: @noindent
5264: may be simplified by optimization into a ``test'' like this:
5265:
5266: @example
5267: (set (cc0) @var{operand})
5268: @end example
5269:
5270: @noindent
5271: So in the machine description, each ``compare'' pattern for an integer
5272: mode must have a corresponding ``test'' pattern that will match the
5273: result of such simplification.
5274:
5275: In some cases machines support instructions identical except for the
5276: machine mode of one or more operands. For example, there may be
5277: ``sign-extend halfword'' and ``sign-extend byte'' instructions whose
5278: patterns are
5279:
5280: @example
5281: (set (match_operand:SI 0 @dots{})
5282: (extend:SI (match_operand:HI 1 @dots{})))
5283:
5284: (set (match_operand:SI 0 @dots{})
5285: (extend:SI (match_operand:QI 1 @dots{})))
5286: @end example
5287:
5288: @noindent
5289: Constant integers do not specify a machine mode, so an instruction to
5290: extend a constant value could match either pattern. The pattern it
5291: actually will match is the one that appears first in the file. For correct
5292: results, this must be the one for the widest possible mode (@code{HImode},
5293: here). If the pattern matches the @code{QImode} instruction, the results
5294: will be incorrect if the constant value does not actually fit that mode.
5295:
5296: Such instructions to extend constants are rarely generated because they are
5297: optimized away, but they do occasionally happen in nonoptimized
5298: compilations.
5299:
5300: @node Jump Patterns, Peephole Definitions, Dependent Patterns, Machine Desc
5301: @section Defining Jump Instruction Patterns
5302:
5303: GNU CC assumes that the machine has a condition code. A comparison insn
5304: sets the condition code, recording the results of both signed and unsigned
5305: comparison of the given operands. A separate branch insn tests the
5306: condition code and branches or not according its value. The branch insns
5307: come in distinct signed and unsigned flavors. Many common machines, such
5308: as the Vax, the 68000 and the 32000, work this way.
5309:
5310: Some machines have distinct signed and unsigned compare instructions, and
5311: only one set of conditional branch instructions. The easiest way to handle
5312: these machines is to treat them just like the others until the final stage
5313: where assembly code is written. At this time, when outputting code for the
5314: compare instruction, peek ahead at the following branch using
5315: @code{NEXT_INSN (insn)}. (The variable @code{insn} refers to the insn
5316: being output, in the output-writing code in an instruction pattern.) If
5317: the RTL says that is an unsigned branch, output an unsigned compare;
5318: otherwise output a signed compare. When the branch itself is output, you
5319: can treat signed and unsigned branches identically.
5320:
5321: The reason you can do this is that GNU CC always generates a pair of
5322: consecutive RTL insns, one to set the condition code and one to test it,
5323: and keeps the pair inviolate until the end.
5324:
5325: To go with this technique, you must define the machine-description macro
5326: @code{NOTICE_UPDATE_CC} to do @code{CC_STATUS_INIT}; in other words, no
5327: compare instruction is superfluous.
5328:
5329: Some machines have compare-and-branch instructions and no condition code.
5330: A similar technique works for them. When it is time to ``output'' a
5331: compare instruction, record its operands in two static variables. When
5332: outputting the branch-on-condition-code instruction that follows, actually
5333: output a compare-and-branch instruction that uses the remembered operands.
5334:
5335: It also works to define patterns for compare-and-branch instructions.
5336: In optimizing compilation, the pair of compare and branch instructions
5337: will be combined accoprding to these patterns. But this does not happen
5338: if optimization is not requested. So you must use one of the solutions
5339: above in addition to any special patterns you define.
5340:
5341: @node Peephole Definitions, Expander Definitions, Jump Patterns, Machine Desc
5342: @section Defining Machine-Specific Peephole Optimizers
5343:
5344: In addition to instruction patterns the @file{md} file may contain
5345: definitions of machine-specific peephole optimizations.
5346:
5347: The combiner does not notice certain peephole optimizations when the data
5348: flow in the program does not suggest that it should try them. For example,
5349: sometimes two consecutive insns related in purpose can be combined even
5350: though the second one does not appear to use a register computed in the
5351: first one. A machine-specific peephole optimizer can detect such
5352: opportunities.
5353:
5354: A definition looks like this:
5355:
5356: @example
5357: (define_peephole
5358: [@var{insn-pattern-1}
5359: @var{insn-pattern-2}
5360: @dots{}]
5361: "@var{condition}"
5362: "@var{template}")
5363: @end example
5364:
5365: In this skeleton, @var{insn-pattern-1} and so on are patterns to match
5366: consecutive instructions. The optimization applies to a sequence of
5367: instructions when @var{insn-pattern-1} matches the first one,
5368: @var{insn-pattern-2} matches the next, and so on.@refill
5369:
5370: @var{insn-pattern-1} and so on look @emph{almost} like the second operand
5371: of @code{define_insn}. There is one important difference: this pattern is
5372: an RTX, not a vector. If the @code{define_insn} pattern would be a vector
5373: of one element, the @var{insn-pattern} should be just that element, no
5374: vector. If the @code{define_insn} pattern would have multiple elements
5375: then the @var{insn-pattern} must place the vector inside an explicit
5376: @code{parallel} RTX.@refill
5377:
5378: The operands of the instructions are matched with @code{match_operands} and
5379: @code{match_dup}, as usual). What is not usual is that the operand numbers
5380: apply to all the instruction patterns in the definition. So, you can check
5381: for identical operands in two instructions by using @code{match_operand}
5382: in one instruction and @code{match_dup} in the other.
5383:
5384: The operand constraints used in @code{match_operand} patterns do not have
5385: any direct effect on the applicability of the optimization, but they will
5386: be validated afterward, so write constraints that are sure to fit whenever
5387: the optimization is applied. It is safe to use @code{"g"} for each
5388: operand.
5389:
5390: Once a sequence of instructions matches the patterns, the @var{condition}
5391: is checked. This is a C expression which makes the final decision whether
5392: to perform the optimization (do so if the expression is nonzero). If
5393: @var{condition} is omitted (in other words, the string is empty) then the
5394: optimization is applied to every sequence of instructions that matches the
5395: patterns.
5396:
5397: The defined peephole optimizations are applied after register allocation is
5398: complete. Therefore, the optimizer can check which operands have ended up
5399: in which kinds of registers, just by looking at the operands.
5400:
5401: The way to refer to the operands in @var{condition} is to write
5402: @code{operands[@var{i}]} for operand number @var{i} (as matched by
5403: @code{(match_operand @var{i} @dots{})}). Use the variable @code{insn} to
5404: refer to the last of the insns being matched; use @code{PREV_INSN} to find
5405: the preceding insns (but be careful to skip over any @samp{note} insns that
5406: intervene).@refill
5407:
5408: When optimizing computations with intermediate results, you can use
5409: @var{condition} to match only when the intermediate results are not used
5410: elsewhere. Use the C expression @code{dead_or_set_p (@var{insn},
5411: @var{op})}, where @var{insn} is the insn in which you expect the value to
5412: be used for the last time (from the value of @code{insn}, together with use
5413: of @code{PREV_INSN}), and @var{op} is the intermediate value (from
5414: @code{operands[@var{i}]}).@refill
5415:
5416: Applying the optimization means replacing the sequence of instructions with
5417: one new instruction. The @var{template} controls ultimate output of
5418: assembler code for this combined instruction. It works exactly like the
5419: template of a @code{define_insn}. Operand numbers in this template are the
5420: same ones used in matching the original sequence of instructions.
5421:
5422: The result of a defined peephole optimizer does not need to match any of
5423: the instruction patterns, and it does not have an opportunity to match
5424: them. The peephole optimizer definition itself serves as the instruction
5425: pattern to control how the instruction is output.
5426:
5427: Defined peephole optimizers are run in the last jump optimization pass, so
5428: the instructions they produce are never combined or rearranged
5429: automatically in any way.
5430:
5431: Here is an example, taken from the 68000 machine description:
5432:
5433: @example
5434: (define_peephole
5435: [(set (reg:SI 15) (plus:SI (reg:SI 15) (const_int 4)))
5436: (set (match_operand:DF 0 "register_operand" "f")
5437: (match_operand:DF 1 "register_operand" "ad"))]
5438: "FP_REG_P (operands[0]) && ! FP_REG_P (operands[1])"
5439: "*
5440: @{
5441: rtx xoperands[2];
5442: xoperands[1] = gen_rtx (REG, SImode, REGNO (operands[1]) + 1);
5443: #ifdef MOTOROLA
5444: output_asm_insn (\"move.l %1,(sp)\", xoperands);
5445: output_asm_insn (\"move.l %1,-(sp)\", operands);
5446: return \"fmove.d (sp)+,%0\";
5447: #else
5448: output_asm_insn (\"movel %1,sp@@\", xoperands);
5449: output_asm_insn (\"movel %1,sp@@-\", operands);
5450: return \"fmoved sp@@+,%0\";
5451: #endif
5452: @}
5453: ")
5454: @end example
5455:
5456: The effect of this optimization is to change
5457:
5458: @example
5459: jbsr _foobar
5460: addql #4,sp
5461: movel d1,sp@@-
5462: movel d0,sp@@-
5463: fmoved sp@@+,fp0
5464: @end example
5465:
5466: @noindent
5467: into
5468:
5469: @example
5470: jbsr _foobar
5471: movel d1,sp@@
5472: movel d0,sp@@-
5473: fmoved sp@@+,fp0
5474: @end example
5475:
5476: @node Expander Definitions,, Peephole Definitions, Machine Desc
5477: @section Defining RTL Sequences for Code Generation
5478:
5479: On some target machines, some standard pattern names for RTL generation
5480: cannot be handled with single insn, but a sequence of RTL insns can
5481: represent them. For these target machines, you can write a
5482: @samp{define_expand} to specify how to generate the sequence of RTL.
5483:
5484: A @samp{define_expand} is an RTL expression that looks almost like a
5485: @samp{define_insn}; but, unlike the latter, a @samp{define_expand} is used
5486: only for RTL generation and it can produce more than one RTL insn.
5487:
5488: A @samp{define_expand} RTX has four operands:
5489:
5490: @itemize @bullet
5491: @item
5492: The name. Each @samp{define_expand} must have a name, since the only
5493: use for it is to refer to it by name.
5494:
5495: @item
5496: The RTL template. This is just like the RTL template for a
5497: @samp{define_peephole} in that it is a vector of RTL expressions
5498: each being one insn.
5499:
5500: @item
5501: The condition, a string containing a C expression. This expression is
5502: used to express how the availability of this pattern depends on
5503: subclasses of target machine, selected by command-line options when
5504: GNU CC is run. This is just like the condition of a
5505: @samp{define_insn} that has a standard name.
5506:
5507: @item
5508: The preparation statements, a string containing zero or more C
5509: statements which are to be executed before RTL code is generated from
5510: the RTL template.
5511:
5512: Usually these statements prepare temporary registers for use as
5513: internal operands in the RTL template, but they can also generate RTL
5514: insns directly by calling routines such as @samp{emit_insn}, etc.
5515: Any such insns precede the ones that come from the RTL template.
5516: @end itemize
5517:
5518: The RTL template, in addition to controlling generation of RTL insns,
5519: also describes the operands that need to be specified when this pattern
5520: is used. In particular, it gives a predicate for each operand.
5521:
5522: A true operand, which need to be specified in order to generate RTL from
5523: the pattern, should be described with a @samp{match_operand} in its first
5524: occurrence in the RTL template. This enters information on the operand's
5525: predicate into the tables that record such things. GNU CC uses the
5526: information to preload the operand into a register if that is required for
5527: valid RTL code. If the operand is referred to more than once, subsequent
5528: references should use @samp{match_dup}.
5529:
5530: The RTL template may also refer to internal ``operands'' which are
5531: temporary registers or labels used only within the sequence made by the
5532: @samp{define_expand}. Internal operands are substituted into the RTL
5533: template with @samp{match_dup}, never with @samp{match_operand}. The
5534: values of the internal operands are not passed in as arguments by the
5535: compiler when it requests use of this pattern. Instead, they are computed
5536: within the pattern, in the preparation statements. These statements
5537: compute the values and store them into the appropriate elements of
5538: @code{operands} so that @samp{match_dup} can find them.
5539:
5540: There are two special macros defined for use in the preparation statements:
5541: @code{DONE} and @code{FAIL}. Use them with a following semicolon,
5542: as a statement.
5543:
5544: @table @code
5545: @item DONE
5546: Use the @code{DONE} macro to end RTL generation for the pattern. The
5547: only RTL insns resulting from the pattern on this occasion will be
5548: those already emitted by explicit calls to @code{emit_insn} within the
5549: preparation statements; the RTL template will not be generated.
5550:
5551: @item FAIL
5552: Make the pattern fail on this occasion. When a pattern fails, it means
5553: that the pattern was not truly available. The calling routines in the
5554: compiler will try other strategies for code generation using other patterns.
5555:
5556: Failure is currently supported only for binary operations (addition,
5557: multiplication, shifting, etc.).
5558:
5559: Do not emit any insns explicitly with @code{emit_insn} before failing.
5560: @end table
5561:
5562: Here is an example, the definition of left-shift for the SPUR chip:
5563:
5564: @example
5565: (define_expand "ashlsi3"
5566: [(set (match_operand:SI 0 "register_operand" "")
5567: (ashift:SI
5568: (match_operand:SI 1 "register_operand" "")
5569: (match_operand:SI 2 "nonmemory_operand" "")))]
5570: ""
5571: "
5572: @{
5573: if (GET_CODE (operands[2]) != CONST_INT
5574: || (unsigned) INTVAL (operands[2]) > 3)
5575: FAIL;
5576: @}")
5577: @end example
5578:
5579: @noindent
5580: This example uses @samp{define_expand} so that it can generate an RTL insn
5581: for shifting when the shift-count is in the supported range of 0 to 3 but
5582: fail in other cases where machine insns aren't available. When it fails,
5583: the compiler tries another strategy using different patterns (such as, a
5584: library call).
5585:
5586: If the compiler were able to handle nontrivial condition-strings in
5587: patterns with names, then there would be possible to use a
5588: @samp{define_insn} in that case. Here is another case (zero-extension on
5589: the 68000) which makes more use of the power of @samp{define_expand}:
5590:
5591: @example
5592: (define_expand "zero_extendhisi2"
5593: [(set (match_operand:SI 0 "general_operand" "")
5594: (const_int 0))
5595: (set (strict_low_part
5596: (subreg:HI
5597: (match_operand:SI 0 "general_operand" "")
5598: 0))
5599: (match_operand:HI 1 "general_operand" ""))]
5600: ""
5601: "operands[1] = make_safe_from (operands[1], operands[0]);")
5602: @end example
5603:
5604: @noindent
5605: Here two RTL insns are generated, one to clear the entire output operand
5606: and the other to copy the input operand into its low half. This sequence
5607: is incorrect if the input operand refers to [the old value of] the output
5608: operand, so the preparation statement makes sure this isn't so. The
5609: function @code{make_safe_from} copies the @code{operands[1]} into a
5610: temporary register if it refers to @code{operands[0]}. It does this
5611: by emitting another RTL insn.
5612:
5613: Finally, a third example shows the use of an internal operand.
5614: Zero-extension on the SPUR chip is done by @samp{and}-ing the result
5615: against a halfword mask. But this mask cannot be represented by a
5616: @samp{const_int} because the constant value is too large to be legitimate
5617: on this machine. So it must be copied into a register with
5618: @code{force_reg} and then the register used in the @samp{and}.
5619:
5620: @example
5621: (define_expand "zero_extendhisi2"
5622: [(set (match_operand:SI 0 "register_operand" "")
5623: (and:SI (subreg:SI
5624: (match_operand:HI 1 "register_operand" "")
5625: 0)
5626: (match_dup 2)))]
5627: ""
5628: "operands[2]
5629: = force_reg (SImode, gen_rtx (CONST_INT,
5630: VOIDmode, 65535)); ")
5631: @end example
5632:
5633: @node Machine Macros, Config, Machine Desc, Top
1.1 root 5634: @chapter Machine Description Macros
5635:
5636: The other half of the machine description is a C header file conventionally
5637: given the name @file{tm-@var{machine}.h}. The file @file{tm.h} should be a
5638: link to it. The header file @file{config.h} includes @file{tm.h} and most
5639: compiler source files include @file{config.h}.
5640:
5641: @menu
1.1.1.2 root 5642: * Run-time Target:: Defining -m options like -m68000 and -m68020.
1.1 root 5643: * Storage Layout:: Defining sizes and alignments of data types.
5644: * Registers:: Naming and describing the hardware registers.
5645: * Register Classes:: Defining the classes of hardware registers.
5646: * Stack Layout:: Defining which way the stack grows and by how much.
1.1.1.2 root 5647: * Library Names:: Specifying names of subroutines to call automatically.
1.1 root 5648: * Addressing Modes:: Defining addressing modes valid for memory operands.
5649: * Condition Code:: Defining how insns update the condition code.
5650: * Assembler Format:: Defining how to write insns and pseudo-ops to output.
5651: * Misc:: Everything else.
5652: @end menu
5653:
5654: @node Run-time Target, Storage Layout, Machine Macros, Machine Macros
5655: @section Run-time Target Specification
5656:
5657: @table @code
5658: @item CPP_PREDEFINES
1.1.1.2 root 5659: Define this to be a string constant containing @samp{-D} options
1.1 root 5660: to define the predefined macros that identify this machine and system.
5661:
5662: For example, on the Sun, one can use the value
5663:
5664: @example
1.1.1.2 root 5665: "-Dmc68000 -Dsun -Dunix"
1.1 root 5666: @end example
5667:
5668: @item extern int target_flags;
5669: This declaration should be present.
5670:
5671: @item TARGET_@dots{}
5672: This series of macros is to allow compiler command arguments to
5673: enable or disable the use of optional features of the target machine.
5674: For example, one machine description serves both the 68000 and
5675: the 68020; a command argument tells the compiler whether it should
5676: use 68020-only instructions or not. This command argument works
5677: by means of a macro @code{TARGET_68020} that tests a bit in
5678: @code{target_flags}.
5679:
5680: Define a macro @code{TARGET_@var{featurename}} for each such option.
5681: Its definition should test a bit in @code{target_flags}; for example:
5682:
5683: @example
5684: #define TARGET_68020 (target_flags & 1)
5685: @end example
5686:
5687: One place where these macros are used is in the condition-expressions
5688: of instruction patterns. Note how @code{TARGET_68020} appears
1.1.1.2 root 5689: frequently in the 68000 machine description file, @file{m68k.md}.
1.1 root 5690: Another place they are used is in the definitions of the other
5691: macros in the @file{tm-@var{machine}.h} file.
5692:
5693: @item TARGET_SWITCHES
1.1.1.2 root 5694: This macro defines names of command options to set and clear
1.1 root 5695: bits in @code{target_flags}. Its definition is an initializer
1.1.1.2 root 5696: with a subgrouping for each command option.
1.1 root 5697:
1.1.1.2 root 5698: Each subgrouping contains a string constant, that defines the option
1.1 root 5699: name, and a number, which contains the bits to set in
5700: @code{target_flags}. A negative number says to clear bits instead;
1.1.1.2 root 5701: the negative of the number is which bits to clear. The actual option
1.1 root 5702: name is made by appending @samp{-m} to the specified name.
5703:
5704: One of the subgroupings should have a null string. The number in
5705: this grouping is the default value for @code{target_flags}. Any
1.1.1.2 root 5706: target options act starting with that value.
1.1 root 5707:
5708: Here is an example which defines @samp{-m68000} and @samp{-m68020}
5709: with opposite meanings, and picks the latter as the default:
5710:
5711: @example
5712: #define TARGET_SWITCHES \
5713: @{ @{ "68020", 1@}, \
5714: @{ "68000", -1@}, \
5715: @{ "", 1@}@}
5716: @end example
5717: @end table
5718:
1.1.1.2 root 5719: Sometimes certain combinations of command options do not make sense on a
5720: particular target machine. You can define a macro @code{OVERRIDE_OPTIONS}
5721: to take account of this. This macro, if defined, is executed once
5722: just after all the command options have been parsed.
5723:
1.1 root 5724: @node Storage Layout, Registers, Run-time Target, Machine Macros
5725: @section Storage Layout
5726:
1.1.1.2 root 5727: Note that the definitions of the macros in this table which are sizes or
5728: alignments measured in bits do not need to be constant. They can be C
5729: expressions that refer to static variables, such as the @code{target_flags}.
5730: @xref{Run-time Target}.
5731:
1.1 root 5732: @table @code
5733: @item BITS_BIG_ENDIAN
5734: Define this macro if the most significant bit in a byte has the lowest
5735: number. This means that bit-field instructions count from the most
5736: significant bit. If the machine has no bit-field instructions, this
5737: macro is irrelevant.
5738:
5739: @item BYTES_BIG_ENDIAN
5740: Define this macro if the most significant byte in a word has the
5741: lowest number.
5742:
5743: @item WORDS_BIG_ENDIAN
1.1.1.2 root 5744: Define this macro if, in a multiword object, the most significant
1.1 root 5745: word has the lowest number.
5746:
5747: @item BITS_PER_UNIT
5748: Number of bits in an addressable storage unit (byte); normally 8.
5749:
5750: @item BITS_PER_WORD
5751: Number of bits in a word; normally 32.
5752:
5753: @item UNITS_PER_WORD
5754: Number of storage units in a word; normally 4.
5755:
5756: @item POINTER_SIZE
5757: Width of a pointer, in bits.
5758:
5759: @item PARM_BOUNDARY
1.1.1.2 root 5760: Alignment required for function parameters on the stack, in bits.
5761:
5762: @item STACK_BOUNDARY
5763: Define this macro if you wish to preserve a certain alignment for
5764: the stack pointer at all times. The definition is a C expression
5765: for the desired alignment (measured in bits).
1.1 root 5766:
5767: @item FUNCTION_BOUNDARY
5768: Alignment required for a function entry point, in bits.
5769:
5770: @item BIGGEST_ALIGNMENT
1.1.1.2 root 5771: Biggest alignment that any data type can require on this machine, in bits.
5772:
5773: @item EMPTY_FIELD_ALIGNMENT
5774: Alignment in bits to be given to a structure bit field that follows an
5775: empty field such as @code{int : 0;}.
5776:
5777: @item STRUCTURE_SIZE_BOUNDARY
5778: Number of bits which any structure or union's size must be a multiple of.
5779: Each structure or union's size is rounded up to a multiple of this.
5780:
5781: If you do not define this macro, the default is the same as
5782: @code{BITS_PER_UNIT}.
1.1 root 5783:
5784: @item STRICT_ALIGNMENT
5785: Define this if instructions will fail to work if given data not
5786: on the nominal alignment. If instructions will merely go slower
5787: in that case, do not define this macro.
1.1.1.4 root 5788:
5789: @item CHECK_FLOAT_VALUE (@var{mode}, @var{value})
5790: A C statement to validate the value @var{value} (or type
5791: @code{double}) for mode @var{mode}. This means that you check whether
5792: @var{value} fits within the possible range of values for mode
5793: @var{mode} on this target machine. The mode @var{mode} is always
5794: @code{SFmode} or @code{DFmode}.
5795:
5796: If @var{value} is not valid, you should call @code{error} to print an
5797: error message and then assign some valid value to @var{value}.
5798: Allowing an invalid value to go through the compiler can produce
5799: incorrect assembler code which may even cause Unix assemblers to
5800: crash.
5801:
5802: This macro need not be defined if there is no work for it to do.
1.1 root 5803: @end table
5804:
5805: @node Registers, Register Classes, Storage Layout, Machine Macros
5806: @section Register Usage
5807:
5808: @table @code
5809: @item FIRST_PSEUDO_REGISTER
5810: Number of hardware registers known to the compiler. They receive
5811: numbers 0 through @code{FIRST_PSEUDO_REGISTER-1}; thus, the first
1.1.1.2 root 5812: pseudo register's number really is assigned the number
1.1 root 5813: @code{FIRST_PSEUDO_REGISTER}.
5814:
5815: @item FIXED_REGISTERS
5816: An initializer that says which registers are used for fixed purposes
5817: all throughout the compiled code and are therefore not available for
1.1.1.2 root 5818: general allocation. These would include the stack pointer, the frame
1.1 root 5819: pointer, the program counter on machines where that is considered one
5820: of the addressable registers, and any other numbered register with a
5821: standard use.
5822:
5823: This information is expressed as a sequence of numbers, separated by
5824: commas and surrounded by braces. The @var{n}th number is 1 if
1.1.1.2 root 5825: register @var{n} is fixed, 0 otherwise.
5826:
5827: The table initialized from this macro, and the table initialized by
5828: the following one, may be overridden at run time either automatically,
5829: by the actions of the macro @code{CONDITIONAL_REGISTER_USAGE}, or by
5830: the user with the command options @samp{-ffixed-@var{reg}},
5831: @samp{-fcall-used-@var{reg}} and @samp{-fcall-saved-@var{reg}}.
1.1 root 5832:
5833: @item CALL_USED_REGISTERS
5834: Like @code{FIXED_REGISTERS} but has 1 for each register that is
5835: clobbered (in general) by function calls as well as for fixed
5836: registers. This macro therefore identifies the registers that are not
5837: available for general allocation of values that must live across
5838: function calls.
5839:
1.1.1.2 root 5840: If a register has 0 in @code{CALL_USED_REGISTERS}, the compiler
1.1 root 5841: automatically saves it on function entry and restores it on function
5842: exit, if the register is used within the function.
5843:
1.1.1.2 root 5844: @item CONDITIONAL_REGISTER_USAGE
5845: Zero or more C statements that may conditionally modify two variables
5846: @code{fixed_regs} and @code{call_used_regs} (both of type @code{char
5847: []}) after they have been initialized from the two preceding macros.
5848:
5849: This is necessary in case the fixed or call-clobbered registers depend
5850: on target flags.
5851:
5852: You need not define this macro if it has no work to do.
5853:
1.1.1.4 root 5854: @item OVERLAPPING_REGNO_P (@var{regno})
5855: If defined, this is a C expression whose value is @var{regno} is
5856: nonzero if hard register number @var{regno} is an overlapping
5857: register. This means a hard register which overlaps a hard register
5858: with a different number. (Such overlap is undesirable, but
5859: occasionally it allows a machine to be supported which otherwise could
5860: not be.) This macro must return nonzero for @emph{all} the registers
5861: which overlap each other. GNU CC can use an overlapping register only
5862: in certain limited ways. It can be used for allocation within a basic
5863: block, and may be spilled for reloading; that is all.
5864:
5865: If this macro is not defined, it means that none of the hard registers
5866: overlap each other. This is the usual situation.
5867:
5868: @item INSN_CLOBBERS_REGNO_P (@var{insn}, @var{regno})
5869: If defined, this is a C expression whose value should be nonzero if
5870: the insn @var{insn} has the effect of mysteriously clobbering the
5871: contents of hard register number @var{regno}. By ``mysterious'' we
5872: mean that the insn's RTL expression doesn't describe such an effect.
5873:
5874: If this macro is not defined, it means that no insn clobbers registers
5875: mysteriously. This is the usual situation; all else being equal,
5876: it is best for the RTL expression to show all the activity.
5877:
5878: @item PRESERVE_DEATH_INFO_REGNO_P (@var{regno})
5879: If defined, this is a C expression whose value is nonzero if accurate
5880: @code{REG_DEAD} notes are needed for hard register number @var{regno}
5881: at the time of outputting the assembler code. When this is so, a few
5882: optimizations that take place after register allocation and could
5883: invalidate the death notes are not done when this register is
5884: involved.
5885:
5886: You would arrange to preserve death info for a register when some
5887: of the code in the machine description which is executed to write
5888: the assembler code looks at the the death notes. This is
5889: necessary only when the actual hardware feature which GNU CC
5890: thinks of as a register is not actually a register of the usual sort.
5891: (It might, for example, be a hardware stack.)
5892:
5893: If this macro is not defined, it means that no death notes need to be
5894: preserved. This is the usual situation.
5895:
1.1 root 5896: @item HARD_REGNO_REGS (@var{regno}, @var{mode})
5897: A C expression for the number of consecutive hard registers, starting
5898: at register number @var{regno}, required to hold a value of mode
5899: @var{mode}.
5900:
5901: On a machine where all registers are exactly one word, a suitable
5902: definition of this macro is
5903:
5904: @example
5905: #define HARD_REGNO_NREGS(REGNO, MODE) \
5906: ((GET_MODE_SIZE (MODE) + UNITS_PER_WORD - 1) \
5907: / UNITS_PER_WORD))
5908: @end example
5909:
5910: @item HARD_REGNO_MODE_OK (@var{regno}, @var{mode})
5911: A C expression that is nonzero if it is permissible to store a value
5912: of mode @var{mode} in hard register number @var{regno} (or in several
5913: registers starting with that one). For a machine where all registers
5914: are equivalent, a suitable definition is
5915:
5916: @example
5917: #define HARD_REGNO_MODE_OK(REGNO, MODE) 1
5918: @end example
5919:
5920: It is not necessary for this macro to check for fixed register numbers
5921: because the allocation mechanism considers them to be always occupied.
5922:
1.1.1.2 root 5923: Many machines have special registers for floating point arithmetic.
5924: Often people assume that floating point machine modes are allowed only
5925: in floating point registers. This is not true. Any registers that
5926: can hold integers can safely @emph{hold} a floating point machine
5927: mode, whether or not floating arithmetic can be done on it in those
5928: registers.
5929:
5930: The true significance of special floating registers is rather than
5931: non-floating-point machine modes @emph{may not} go in those registers.
5932: This is true if the floating registers normalize any value stored in
5933: them, because storing a non-floating value there would garble it. If
5934: the floating registers do not automatically normalize, if you can
5935: store any bit pattern in one and retrieve it unchanged without a trap,
5936: then any machine mode may go in a floating register and this macro
5937: should say so.
5938:
5939: Sometimes there are floating registers that are especially slow to
5940: access, so that it is better to store a value in a stack frame than in
5941: such a register if floating point arithmetic is not being done. As long
5942: as the floating registers are not in class @code{GENERAL_REGS}, they
5943: will not be used unless some insn's constraint asks for one.
5944:
5945: It is obligatory to support floating point `move' instructions into
5946: and out of general registers, because unions and structures (which
5947: have modes @samp{SImode} or @samp{DImode}) can be in those registers
5948: and they may have floating point members.
5949:
1.1 root 5950: @item MODES_TIEABLE_P (@var{mode1}, @var{mode2})
5951: A C expression that is nonzero if it is desirable to choose register
5952: allocation so as to avoid move instructions between a value of mode
5953: @var{mode1} and a value of mode @var{mode2}.
5954:
5955: If @code{HARD_REGNO_MODE_OK (@var{r}, @var{mode1})} and
5956: @code{HARD_REGNO_MODE_OK (@var{r}, @var{mode2})} are ever different
5957: for any @var{r}, then @code{MODES_TIEABLE_P (@var{mode1},
5958: @var{mode2})} must be zero.
5959:
5960: @item PC_REGNUM
5961: If the program counter has a register number, define this as that
5962: register number. Otherwise, do not define it.
5963:
5964: @item STACK_POINTER_REGNUM
5965: The register number of the stack pointer register, which must also be
5966: a fixed register according to @code{FIXED_REGISTERS}. On many
5967: machines, the hardware determines which register this is.
5968:
5969: @item FRAME_POINTER_REGNUM
5970: The register number of the frame pointer register, which is used to
1.1.1.2 root 5971: access automatic variables in the stack frame. On some machines, the
1.1 root 5972: hardware determines which register this is. On other machines, you
5973: can choose any register you wish for this purpose.
5974:
1.1.1.2 root 5975: @item FRAME_POINTER_REQUIRED
5976: A C expression which is nonzero if a function must have and use a
5977: frame pointer. This expression is evaluated in the reload pass, in
5978: the function @code{reload}, and it can in principle examine the
5979: current function and decide according to the facts, but on most
5980: machines the constant 0 or the constant 1 suffices. Use 0 when the
5981: machine allows code to be generated with no frame pointer, and doing
5982: so saves some time or space. Use 1 when there is no possible
5983: advantage to avoiding a frame pointer.
5984:
5985: In certain cases, the compiler does not know how to do without a frame
5986: pointer. The compiler recognizes those cases and automatically gives
5987: the function a frame pointer regardless of what
5988: @code{FRAME_POINTER_REQUIRED} says. You don't need to worry about
5989: them.@refill
5990:
5991: In a function that does not require a frame pointer, the frame pointer
1.1.1.4 root 5992: register can be allocated for ordinary usage, unless you mark it as a
5993: fixed register. See @code{FIXED_REGISTERS} for more information.
1.1.1.2 root 5994:
1.1 root 5995: @item ARG_POINTER_REGNUM
5996: The register number of the arg pointer register, which is used to
5997: access the function's argument list. On some machines, this is the
5998: same as the frame pointer register. On some machines, the hardware
5999: determines which register this is. On other machines, you can choose
1.1.1.4 root 6000: any register you wish for this purpose. If this is not the same
6001: register as the frame pointer register, then you must mark it as a
1.1 root 6002: fixed register according to @code{FIXED_REGISTERS}.
6003:
6004: @item STATIC_CHAIN_REGNUM
6005: The register number used for passing a function's static chain
6006: pointer. This is needed for languages such as Pascal and Algol where
6007: functions defined within other functions can access the local
6008: variables of the outer functions; it is not currently used because C
6009: does not provide this feature.
6010:
6011: The static chain register need not be a fixed register.
6012:
6013: @item STRUCT_VALUE_REGNUM
1.1.1.2 root 6014: When a function's value's mode is @code{BLKmode}, the value is not
6015: returned according to @code{FUNCTION_VALUE}. Instead, the caller
6016: passes the address of a block of memory in which the value should be
6017: stored. @code{STRUCT_VALUE_REGNUM} is the register in which this
6018: address is passed.
1.1.1.5 ! root 6019:
! 6020: @item REG_ALLOC_ORDER
! 6021: If defined, an initializer for a vector of integers, containing the
! 6022: numbers of hard registers in the order in which the GNU CC should
! 6023: prefer to use them (from most preferred to least).
! 6024:
! 6025: If this macro is not defined, registers are used lowest numbered first
! 6026: (all else being equal).
! 6027:
! 6028: One use of this macro is on the 360, where the highest numbered
! 6029: registers must always be saved and the save-multiple-registers
! 6030: instruction supports only sequences of consecutive registers. This
! 6031: macro is defined to cause the highest numbered allocatable registers
! 6032: to be used first.
1.1 root 6033: @end table
6034:
6035: @node Register Classes, Stack Layout, Registers, Machine Macros
6036: @section Register Classes
6037:
6038: On many machines, the numbered registers are not all equivalent.
6039: For example, certain registers may not be allowed for indexed addressing;
6040: certain registers may not be allowed in some instructions. These machine
6041: restrictions are described to the compiler using @dfn{register classes}.
6042:
6043: You define a number of register classes, giving each one a name and saying
6044: which of the registers belong to it. Then you can specify register classes
6045: that are allowed as operands to particular instruction patterns.
6046:
6047: In general, each register will belong to several classes. In fact, one
6048: class must be named @code{ALL_REGS} and contain all the registers. Another
6049: class must be named @code{NO_REGS} and contain no registers. Often the
6050: union of two classes will be another class; however, this is not required.
6051:
6052: One of the classes must be named @code{GENERAL_REGS}. There is nothing
6053: terribly special about the name, but the operand constraint letters
6054: @samp{r} and @samp{g} specify this class. If @code{GENERAL_REGS} is
6055: the same as @code{ALL_REGS}, just define it as a macro which expands
6056: to @code{ALL_REGS}.
6057:
6058: The way classes other than @code{GENERAL_REGS} are specified in operand
6059: constraints is through machine-dependent operand constraint letters.
6060: You can define such letters to correspond to various classes, then use
6061: them in operand constraints.
6062:
1.1.1.2 root 6063: You should define a class for the union of two classes whenever some
6064: instruction allows both classes. For example, if an instruction allows
6065: either a floating-point (coprocessor) register or a general register for a
6066: certain operand, you should define a class @code{FLOAT_OR_GENERAL_REGS}
6067: which includes both of them. Otherwise you will get suboptimal code.
6068:
1.1 root 6069: You must also specify certain redundant information about the register
6070: classes: for each class, which classes contain it and which ones are
6071: contained in it; for each pair of classes, the largest class contained
6072: in their union.
6073:
6074: @table @code
6075: @item enum reg_class
6076: An enumeral type that must be defined with all the register class names
6077: as enumeral values. @code{NO_REGS} must be first. @code{ALL_REGS}
6078: must be the last register class, followed by one more enumeral value,
6079: @code{LIM_REG_CLASSES}, which is not a register class but rather
6080: tells how many classes there are.
6081:
6082: Each register class has a number, which is the value of casting
6083: the class name to type @code{int}. The number serves as an index
6084: in many of the tables described below.
6085:
6086: @item REG_CLASS_NAMES
6087: An initializer containing the names of the register classes as C string
6088: constants. These names are used in writing some of the debugging dumps.
6089:
6090: @item REG_CLASS_CONTENTS
6091: An initializer containing the contents of the register classes, as integers
6092: which are bit masks. The @var{n}th integer specifies the contents of class
6093: @var{n}. The way the integer @var{mask} is interpreted is that
6094: register @var{r} is in the class if @code{@var{mask} & (1 << @var{r})} is 1.
6095:
6096: When the machine has more than 32 registers, an integer does not suffice.
6097: Then the integers are replaced by sub-initializers, braced groupings containing
6098: several integers. Each sub-initializer must be suitable as an initializer
6099: for the type @code{HARD_REG_SET} which is defined in @file{hard-reg-set.h}.
6100:
6101: @item REGNO_REG_CLASS (@var{regno})
6102: A C expression whose value is a register class containing hard register
6103: @var{regno}. In general there is more that one such class; choose a class
6104: which is @dfn{minimal}, meaning that no smaller class also contains the
6105: register.
6106:
6107: @item INDEX_REG_CLASS
6108: A macro whose definition is the name of the class to which a valid index
6109: register must belong.
6110:
6111: @item REG_CLASS_FROM_LETTER (@var{char})
6112: A C expression which defines the machine-dependent operand constraint
6113: letters for register classes. If @var{char} is such a letter, the value
6114: should be the register class corresponding to it. Otherwise, the value
6115: should be @code{NO_REGS}.
6116:
1.1.1.2 root 6117: @item REGNO_OK_FOR_BASE_P (@var{num})
6118: A C expression which is nonzero if register number @var{num} is
6119: suitable for use as a base register in operand addresses. It may be
6120: either a suitable hard register or a pseudo register that has been
6121: allocated such a hard register.
6122:
6123: @item REGNO_OK_FOR_INDEX_P (@var{num})
6124: A C expression which is nonzero if register number @var{num} is
6125: suitable for use as an index register in operand addresses. It may be
6126: either a suitable hard register or a pseudo register that has been
6127: allocated such a hard register.
6128:
6129: The difference between an index register and a base register is that
6130: the index register may be scaled. If an address involves the sum of
6131: two registers, neither one of them scaled, then either one may be
6132: labeled the ``base'' and the other the ``index''; but whichever
6133: labeling is used must fit the machine's constraints of which registers
6134: may serve in each capacity. The compiler will try both labelings,
6135: looking for one that is valid, and reload one or both registers only
6136: if neither labeling works.
1.1 root 6137:
6138: @item PREFERRED_RELOAD_CLASS (@var{x}, @var{class})
6139: A C expression that places additional restrictions on the register class
6140: to use when it is necessary to copy value @var{x} into a register in class
6141: @var{class}. The value is a register class; perhaps @var{class}, or perhaps
6142: another, smaller class. @var{class} is always safe as a value. In fact,
6143: the definition
6144:
6145: @example
6146: #define PREFERRED_RELOAD_CLASS(X,CLASS) CLASS
6147: @end example
6148:
6149: @noindent
6150: is always safe. However, sometimes returning a more restrictive class
6151: makes better code. For example, on the 68000, when @var{x} is an
6152: integer constant that is in range for a @samp{moveq} instruction,
6153: the value of this macro is always @code{DATA_REGS} as long as
6154: @var{class} includes the data registers. Requiring a data register
6155: guarantees that a @samp{moveq} will be used.
1.1.1.2 root 6156:
6157: @item CLASS_MAX_NREGS (@var{class}, @var{mode})
6158: A C expression for the maximum number of consecutive registers
6159: of class @var{class} needed to hold a value of mode @var{mode}.
6160:
6161: This is closely related to the macro @code{HARD_REGNO_NREGS}.
6162: In fact, the value of the macro @code{CLASS_MAX_NREGS (@var{class}, @var{mode})}
6163: should be the maximum value of @code{HARD_REGNO_NREGS (@var{regno}, @var{mode})}
6164: for all @var{regno} values in the class @var{class}.
6165:
6166: This macro helps control the handling of multiple-word values
6167: in the reload pass.
1.1 root 6168: @end table
6169:
1.1.1.2 root 6170: Two other special macros describe which constants fit which constraint
6171: letters.
1.1 root 6172:
6173: @table @code
6174: @item CONST_OK_FOR_LETTER_P (@var{value}, @var{c})
6175: A C expression that defines the machine-dependent operand constraint letters
6176: that specify particular ranges of integer values. If @var{c} is one
6177: of those letters, the expression should check that @var{value}, an integer,
6178: is in the appropriate range and return 1 if so, 0 otherwise. If @var{c} is
6179: not one of those letters, the value should be 0 regardless of @var{value}.
6180:
6181: @item CONST_DOUBLE_OK_FOR_LETTER_P (@var{value}, @var{c})
6182: A C expression that defines the machine-dependent operand constraint
6183: letters that specify particular ranges of floating values. If @var{c} is
1.1.1.2 root 6184: one of those letters, the expression should check that @var{value}, an RTX
1.1 root 6185: of code @samp{const_double}, is in the appropriate range and return 1 if
6186: so, 0 otherwise. If @var{c} is not one of those letters, the value should
6187: be 0 regardless of @var{value}.
6188: @end table
6189:
1.1.1.2 root 6190: @node Stack Layout, Library Names, Register Classes, Machine Macros
1.1 root 6191: @section Describing Stack Layout
6192:
6193: @table @code
6194: @item STACK_GROWS_DOWNWARD
6195: Define this macro if pushing a word onto the stack moves the stack
1.1.1.2 root 6196: pointer to a smaller address.
6197:
6198: When we say, ``define this macro if @dots{},'' it means that the
6199: compiler checks this macro only with @code{#ifdef} so the precise
6200: definition used does not matter.
1.1 root 6201:
6202: @item FRAME_GROWS_DOWNWARD
6203: Define this macro if the addresses of local variable slots are at negative
6204: offsets from the frame pointer.
6205:
6206: @item STARTING_FRAME_OFFSET
6207: Offset from the frame pointer to the first local variable slot to be allocated.
6208:
6209: If @code{FRAME_GROWS_DOWNWARD}, the next slot's offset is found by
6210: subtracting the length of the first slot from @code{STARTING_FRAME_OFFSET}.
6211: Otherwise, it is found by adding the length of the first slot to
6212: the value @code{STARTING_FRAME_OFFSET}.
6213:
6214: @item PUSH_ROUNDING (@var{npushed})
6215: A C expression that is the number of bytes actually pushed onto the
6216: stack when an instruction attempts to push @var{npushed} bytes.
6217:
1.1.1.2 root 6218: If the target machine does not have a push instruction, do not define
6219: this macro. That directs GNU CC to use an alternate strategy: to
6220: allocate the entire argument block and then store the arguments into
6221: it.
6222:
1.1 root 6223: On some machines, the definition
6224:
6225: @example
6226: #define PUSH_ROUNDING(BYTES) (BYTES)
6227: @end example
6228:
6229: @noindent
6230: will suffice. But on other machines, instructions that appear
6231: to push one byte actually push two bytes in an attempt to maintain
6232: alignment. Then the definition should be
6233:
6234: @example
6235: #define PUSH_ROUNDING(BYTES) (((BYTES) + 1) & ~1)
6236: @end example
6237:
6238: @item FIRST_PARM_OFFSET
6239: Offset from the argument pointer register to the first argument's address.
6240:
1.1.1.2 root 6241: @item RETURN_POPS_ARGS (@var{funtype})
6242: A C expression that should be 1 if a function pops its own arguments
6243: on returning, or 0 if the function pops no arguments and the caller
6244: must therefore pop them all after the function returns.
6245:
6246: @var{funtype} is a C variable whose value is a tree node that
6247: describes the function in question. Normally it is a node of type
6248: @code{FUNCTION_TYPE} that describes the data type of the function.
6249: From this it is possible to obtain the data types of the value and
6250: arguments (if known).
6251:
6252: When a call to a library function is being considered, @var{funtype}
6253: will contain an identifier node for the library function. Thus, if
6254: you need to distinguish among various library functions, you can do so
6255: by their names. Note that ``library function'' in this context means
6256: a function used to perform arithmetic, whose name is known specially
6257: in the compiler and was not mentioned in the C code being compiled.
6258:
6259: On the Vax, all functions always pop their arguments, so the
6260: definition of this macro is 1. On the 68000, using the standard
6261: calling convention, no functions pop their arguments, so the value of
6262: the macro is always 0 in this case. But an alternative calling
6263: convention is available in which functions that take a fixed number of
6264: arguments pop them but other functions (such as @code{printf}) pop
6265: nothing (the caller pops all). When this convention is in use,
6266: @var{funtype} is examined to determine whether a function takes a
6267: fixed number of arguments.
6268:
6269: @item FUNCTION_VALUE (@var{valtype}, @var{func})
6270: A C expression to create an RTX representing the place where a
6271: function returns a value of data type @var{valtype}. @var{valtype} is
6272: a tree node representing a data type. Write @code{TYPE_MODE
6273: (@var{valtype})} to get the machine mode used to represent that type.
6274: On many machines, only the mode is relevant. (Actually, on most
6275: machines, scalar values are returned in the same place regardless of
6276: mode).@refill
6277:
6278: If the precise function being called is known, @var{func} is a tree
6279: node (@code{FUNCTION_DECL}) for it; otherwise, @var{func} is a null
6280: pointer. This makes it possible to use a different value-returning
6281: convention for specific functions when all their calls are
6282: known.@refill
6283:
6284: @item FUNCTION_OUTGOING_VALUE (@var{valtype}, @var{func})
6285: Define this macro if the target machine has ``register windows''
6286: so that the register in which a function returns its value is not
6287: the same as the one in which the caller sees the value.
6288:
6289: For such machines, @code{FUNCTION_VALUE} computes the register in
6290: which the caller will see the value, and
6291: @code{FUNCTION_OUTGOING_VALUE} should be defined in a similar fashion
6292: to tell the function where to put the value.@refill
6293:
6294: If @code{FUNCTION_OUTGOING_VALUE} is not defined,
6295: @code{FUNCTION_VALUE} serves both purposes.@refill
6296:
6297: @item LIBCALL_VALUE (@var{mode})
6298: A C expression to create an RTX representing the place where a library
6299: function returns a value of mode @var{mode}. If the precise function
6300: being called is known, @var{func} is a tree node
6301: (@code{FUNCTION_DECL}) for it; otherwise, @var{func} is a null
6302: pointer. This makes it possible to use a different value-returning
6303: convention for specific functions when all their calls are
6304: known.@refill
6305:
6306: Note that ``library function'' in this context means a compiler
6307: support routine, used to perform arithmetic, whose name is known
6308: specially by the compiler and was not mentioned in the C code being
6309: compiled.
6310:
6311: @item FUNCTION_VALUE_REGNO_P (@var{regno})
6312: A C expression that is nonzero if @var{regno} is the number of a hard
6313: register in which function values are sometimes returned.
6314:
6315: A register whose use for returning values is limited to serving as the
6316: second of a pair (for a value of type @code{double}, say) need not be
6317: recognized by this macro. So for most machines, this definition
6318: suffices:
6319:
6320: @example
6321: #define FUNCTION_VALUE_REGNO_P(N) ((N) == 0)
6322: @end example
6323:
6324: @item FUNCTION_ARG (@var{cum}, @var{mode}, @var{type}, @var{named})
6325: A C expression that controls whether a function argument is passed
6326: in a register, and which register.
6327:
6328: The arguments are @var{cum}, which summarizes all the previous
6329: arguments; @var{mode}, the machine mode of the argument; @var{type},
6330: the data type of the argument as a tree node or 0 if that is not known
6331: (which happens for C support library functions); and @var{named},
6332: which is 1 for an ordinary argument and 0 for nameless arguments that
6333: correspond to @samp{...} in the called function's prototype.
6334:
6335: The value of the expression should either be a @samp{reg} RTX for the
6336: hard register in which to pass the argument, or zero to pass the
6337: argument on the stack.
6338:
6339: For the Vax and 68000, where normally all arguments are pushed, zero
6340: suffices as a definition.
6341:
6342: @item FUNCTION_INCOMING_ARG (@var{cum}, @var{mode}, @var{type}, @var{named})
6343: Define this macro if the target machine has ``register windows'', so
6344: that the register in which a function sees an arguments is not
6345: necessarily the same as the one in which the caller passed the
6346: argument.
6347:
6348: For such machines, @code{FUNCTION_ARG} computes the register in which
6349: the caller passes the value, and @code{FUNCTION_INCOMING_ARG} should
6350: be defined in a similar fashion to tell the function being called
6351: where the arguments will arrive.
6352:
6353: If @code{FUNCTION_INCOMING_ARG} is not defined, @code{FUNCTION_ARG}
6354: serves both purposes.@refill
6355:
6356: @item FUNCTION_ARG_PARTIAL_NREGS (@var{cum}, @var{mode}, @var{type}, @var{named})
6357: A C expression for the number of words, at the beginning of an
6358: argument, must be put in registers. The value must be zero for
6359: arguments that are passed entirely in registers or that are entirely
6360: pushed on the stack.
6361:
6362: On some machines, certain arguments must be passed partially in
6363: registers and partially in memory. On these machines, typically the
6364: first @var{n} words of arguments are passed in registers, and the rest
6365: on the stack. If a multi-word argument (a @code{double} or a
6366: structure) crosses that boundary, its first few words must be passed
6367: in registers and the rest must be pushed. This macro tells the
6368: compiler when this occurs, and how many of the words should go in
6369: registers.
6370:
6371: @code{FUNCTION_ARG} for these arguments should return the first
6372: register to be used by the caller for this argument; likewise
6373: @code{FUNCTION_INCOMING_ARG}, for the called function.
6374:
6375: @item CUMULATIVE_ARGS
6376: A C type for declaring a variable that is used as the first argument
6377: of @code{FUNCTION_ARG} and other related values. For some target
6378: machines, the type @code{int} suffices and can hold the number of
6379: bytes of argument so far.
6380:
6381: @item INIT_CUMULATIVE_ARGS (@var{cum})
6382: A C statement (sans semicolon) for initializing the variable @var{cum}
6383: for the state at the beginning of the argument list. The variable has
6384: type @code{CUMULATIVE_ARGS}.
6385:
6386: @item FUNCTION_ARG_ADVANCE (@var{cum}, @var{mode}, @var{type}, @var{named})
6387: Update the summarizer variable @var{cum} to advance past an argument
6388: in the argument list. The values @var{mode}, @var{type} and
6389: @var{named} describe that argument. Once this is done, the variable
6390: @var{cum} is suitable for analyzing the @emph{following} argument
6391: with @code{FUNCTION_ARG}, etc.@refill
6392:
6393: @item FUNCTION_ARG_REGNO_P (@var{regno})
6394: A C expression that is nonzero if @var{regno} is the number of a hard
6395: register in which function arguments are sometimes passed. This does
6396: @emph{not} include implicit arguments such as the static chain and
6397: the structure-value address. On many machines, no registers can be
6398: used for this purpose since all function arguments are pushed on the
6399: stack.
1.1 root 6400:
6401: @item FUNCTION_PROLOGUE (@var{file}, @var{size})
6402: A C compound statement that outputs the assembler code for entry to a
6403: function. The prologue is responsible for setting up the stack frame,
6404: initializing the frame pointer register, saving registers that must be
1.1.1.2 root 6405: saved, and allocating @var{size} additional bytes of storage for the
6406: local variables. @var{size} is an integer. @var{file} is a stdio
6407: stream to which the assembler code should be output.
1.1 root 6408:
6409: The label for the beginning of the function need not be output by this
6410: macro. That has already been done when the macro is run.
6411:
6412: To determine which registers to save, the macro can refer to the array
1.1.1.2 root 6413: @code{regs_ever_live}: element @var{r} is nonzero if hard register
6414: @var{r} is used anywhere within the function. This implies the
6415: function prologue should save register @var{r}, but not if it is one
6416: of the call-used registers.
6417:
6418: On machines where functions may or may not have frame-pointers, the
6419: function entry code must vary accordingly; it must set up the frame
6420: pointer if one is wanted, and not otherwise. To determine whether a
6421: frame pointer is in wanted, the macro can refer to the variable
6422: @code{frame_pointer_needed}. The variable's value will be 1 at run
6423: time in a function that needs a frame pointer.
6424:
6425: @item FUNCTION_PROFILER (@var{file}, @var{labelno})
6426: A C statement or compound statement to output to @var{file} some
6427: assembler code to call the profiling subroutine @code{mcount}.
6428: Before calling, the assembler code must load the address of a
6429: counter variable into a register where @code{mcount} expects to
6430: find the address. The name of this variable is @samp{LP} followed
6431: by the number @var{labelno}, so you would generate the name using
6432: @samp{LP%d} in a @code{fprintf}.
6433:
6434: The details of how the address should be passed to @code{mcount} are
6435: determined by your operating system environment, not by GNU CC. To
6436: figure them out, compile a small program for profiling using the
6437: system's installed C compiler and look at the assembler code that
6438: results.
6439:
6440: @item EXIT_IGNORES_STACK
6441: Define this macro as a C expression that is nonzero if the return
6442: instruction or the function epilogue ignores the value of the stack
6443: pointer; in other words, if it is safe to delete an instruction to
6444: adjust the stack pointer before a return from the function.
6445:
6446: Note that this macro's value is relevant only for for which frame
6447: pointers are maintained. It is never possible to delete a final stack
6448: adjustment in a function that has no frame pointer, and the compiler
6449: knows this regardless of @code{EXIT_IGNORES_STACK}.
1.1 root 6450:
6451: @item FUNCTION_EPILOGUE (@var{file}, @var{size})
6452: A C compound statement that outputs the assembler code for exit from a
6453: function. The epilogue is responsible for restoring the saved
6454: registers and stack pointer to their values when the function was
6455: called, and returning control to the caller. This macro takes the
6456: same arguments as the macro @code{FUNCTION_PROLOGUE}, and the
6457: registers to restore are determined from @code{regs_ever_live} and
6458: @code{CALL_USED_REGISTERS} in the same way.
6459:
1.1.1.2 root 6460: On some machines, there is a single instruction that does all the work
6461: of returning from the function. On these machines, give that
6462: instruction the name @samp{return} and do not define the macro
6463: @code{FUNCTION_EPILOGUE} at all.
6464:
6465: On machines where functions may or may not have frame-pointers, the
6466: function exit code must vary accordingly. Sometimes the code for
6467: these two cases is completely different. To determine whether a frame
6468: pointer is in wanted, the macro can refer to the variable
6469: @code{frame_pointer_needed}. The variable's value will be 1 at run
6470: time in a function that needs a frame pointer.
6471:
6472: On some machines, some functions pop their arguments on exit while
6473: others leave that for the caller to do. For example, the 68020 when
6474: given @samp{-mrtd} pops arguments in functions that take a fixed
6475: number of arguments.
6476:
6477: Your definition of the macro @code{RETURN_POPS_ARGS} decides which
6478: functions pop their own arguments. @code{FUNCTION_EPILOGUE} needs to
6479: know what was decided. The variable @code{current_function_pops_args}
6480: is nonzero if the function should pop its own arguments. If so, use
6481: the variable @code{current_function_args_size} as the number of bytes
6482: to pop.
6483:
6484: @item FIX_FRAME_POINTER_ADDRESS (@var{addr}, @var{depth})
6485: A C compound statement to alter a memory address that uses the frame
6486: pointer register so that it uses the stack pointer register instead.
6487: This must be done in the instructions that load parameter values into
6488: registers, when the reload pass determines that a frame pointer is not
6489: necessary for the function. @var{addr} will be a C variable name, and
6490: the updated address should be stored in that variable. @var{depth}
6491: will be the current depth of stack temporaries (number of bytes of
6492: arguments currently pushed). The change in offset between a
6493: frame-pointer-relative address and a stack-pointer-relative address
6494: must include @var{depth}.
6495:
6496: Even if your machine description specifies there will always be a
6497: frame pointer in the frame pointer register, you must still define
6498: @code{FIX_FRAME_POINTER_ADDRESS}, but the definition will never be
6499: executed at run time, so it may be empty.
6500: @end table
6501:
6502: @node Library Names, Addressing Modes, Stack Layout, Machine Macros
6503: @section Library Subroutine Names
6504:
6505: @table @code
6506: @item UDIVSI3_LIBCALL
6507: A C string constant giving the name of the function to call for
6508: division of a full-word by a full-word. If you do not define this
6509: macro, the default name is used, which is @code{_udivsi3}, a function
6510: defined in @file{gnulib}.
6511:
6512: @item UMODSI3_LIBCALL
6513: A C string constant giving the name of the function to call for the
6514: remainder in division of a full-word by a full-word. If you do not
6515: define this macro, the default name is used, which is @code{_umodsi3},
6516: a function defined in @file{gnulib}.
6517:
6518: @item TARGET_MEM_FUNCTIONS
6519: Define this macro if GNU CC should generate calls to the System V
6520: (and ANSI C) library functions @code{memcpy} and @code{memset}
6521: rather than the BSD functions @code{bcopy} and @code{bzero}.
1.1 root 6522: @end table
6523:
1.1.1.2 root 6524: @node Addressing Modes, Misc, Library Names, Machine Macros
1.1 root 6525: @section Addressing Modes
6526:
6527: @table @code
6528: @item HAVE_POST_INCREMENT
6529: Define this macro if the machine supports post-increment addressing.
6530:
6531: @item HAVE_PRE_INCREMENT
6532: @itemx HAVE_POST_DECREMENT
6533: @itemx HAVE_PRE_DECREMENT
6534: Similar for other kinds of addressing.
6535:
6536: @item CONSTANT_ADDRESS_P (@var{x})
1.1.1.2 root 6537: A C expression that is 1 if the RTX @var{x} is a constant whose value
1.1 root 6538: is an integer. This includes integers whose values are not explicitly
1.1.1.2 root 6539: known, such as @samp{symbol_ref} and @samp{label_ref} expressions and
6540: @samp{const} arithmetic expressions.
6541:
6542: On most machines, this can be defined as @code{CONSTANT_P (@var{x})},
6543: but a few machines are more restrictive in which constant addresses
6544: are supported.
1.1 root 6545:
6546: @item MAX_REGS_PER_ADDRESS
6547: A number, the maximum number of registers that can appear in a valid
6548: memory address.
6549:
6550: @item GO_IF_LEGITIMATE_ADDRESS (@var{mode}, @var{x}, @var{label})
6551: A C compound statement with a conditional @code{goto @var{label};}
1.1.1.2 root 6552: executed if @var{x} (an RTX) is a legitimate memory address on the
6553: target machine for a memory operand of mode @var{mode}.
1.1 root 6554:
6555: It usually pays to define several simpler macros to serve as
1.1.1.2 root 6556: subroutines for this one. Otherwise it may be too complicated to
6557: understand.
6558:
6559: This macro must exist in two variants: a strict variant and a
6560: non-strict one. The strict variant is used in the reload pass. It
6561: must be defined so that any pseudo-register that has not been
6562: allocated a hard register is considered a memory reference. In
6563: contexts where some kind of register is required, a pseudo-register
6564: with no hard register must be rejected.
6565:
6566: The non-strict variant is used in other passes. It must be defined to
6567: accept all pseudo-registers in every context where some kind of
6568: register is required.
6569:
6570: Compiler source files that want to use the strict variant of this
6571: macro define the macro @code{REG_OK_STRICT}. You should use an
6572: @code{#ifdef REG_OK_STRICT} conditional to define the strict variant
6573: in that case and the non-strict variant otherwise.
6574:
6575: Typically among the subroutines used to define
6576: @code{GO_IF_LEGITIMATE_ADDRESS} are subroutines to check for
6577: acceptable registers for various purposes (one for base registers, one
6578: for index registers, and so on). Then only these subroutine macros
6579: need have two variants; the higher levels of macros may be the same
6580: whether strict or not.@refill
1.1 root 6581:
6582: @item LEGITIMIZE_ADDRESS (@var{x}, @var{oldx}, @var{mode}, @var{win})
6583: A C compound statement that attempts to replace @var{x} with a valid
1.1.1.2 root 6584: memory address for an operand of mode @var{mode}. @var{win} will be a
6585: C statement label elsewhere in the code; the macro definition may use
1.1 root 6586:
6587: @example
6588: GO_IF_LEGITIMATE_ADDRESS (@var{mode}, @var{x}, @var{win});
6589: @end example
6590:
6591: @noindent
6592: to avoid further processing if the address has become legitimate.
6593:
6594: @var{x} will always be the result of a call to @code{break_out_memory_refs},
6595: and @var{oldx} will be the operand that was given to that function to produce
6596: @var{x}.
6597:
1.1.1.2 root 6598: The code generated by this macro should not alter the substructure of
6599: @var{x}. If it transforms @var{x} into a more legitimate form, it
6600: should assign @var{x} (which will always be a C variable) a new value.
6601:
6602: It is not necessary for this macro to come up with a legitimate
6603: address. The compiler has standard ways of doing so in all cases. In
6604: fact, it is safe for this macro to do nothing. But often a
6605: machine-dependent strategy can generate better code.
6606:
6607: @item GO_IF_MODE_DEPENDENT_ADDRESS (@var{addr}, @var{label})
6608: A C statement or compound statement with a conditional @code{goto
6609: @var{label};} executed if memory address @var{x} (an RTX) can have
6610: different meanings depending on the machine mode of the memory
6611: reference it is used for.
6612:
6613: Autoincrement and autodecrement addresses typically have mode-dependent
6614: effects because the amount of the increment or decrement is the size
6615: of the operand being addressed. Some machines have other mode-dependent
6616: addresses. Many RISC machines have no mode-dependent addresses.
6617:
6618: You may assume that @var{addr} is a valid address for the machine.
6619:
6620: @item LEGITIMATE_CONSTANT_P (@var{x})
6621: A C expression that is nonzero if @var{x} is a legitimate constant for
6622: an immediate operand on the target machine. You can assume that
6623: either @var{x} is a @samp{const_double} or it satisfies
6624: @code{CONSTANT_P}, so you need not check these things. In fact,
6625: @samp{1} is a suitable definition for this macro on machines where any
6626: @samp{const_double} is valid and anything @code{CONSTANT_P} is valid.@refill
1.1 root 6627: @end table
6628:
6629: @node Misc, Condition Code, Addressing Modes, Machine Macros
6630: @section Miscellaneous Parameters
6631:
6632: @table @code
6633: @item CASE_VECTOR_MODE
1.1.1.2 root 6634: An alias for a machine mode name. This is the machine mode that
6635: elements of a jump-table should have.
1.1 root 6636:
6637: @item CASE_VECTOR_PC_RELATIVE
6638: Define this macro if jump-tables should contain relative addresses.
6639:
1.1.1.2 root 6640: @item CASE_DROPS_THROUGH
6641: Define this if control falls through a @code{case} insn when the index
6642: value is out of range. This means the specified default-label is
6643: actually ignored by the @code{case} insn proper.
6644:
1.1 root 6645: @item IMPLICIT_FIX_EXPR
6646: An alias for a tree code that should be used by default for conversion
1.1.1.2 root 6647: of floating point values to fixed point. Normally,
6648: @code{FIX_ROUND_EXPR} is used.@refill
6649:
6650: @item FIXUNS_TRUNC_LIKE_FIX_TRUNC
6651: Define this macro if the same instructions that convert a floating
6652: point number to a signed fixed point number also convert validly to an
6653: unsigned one.
1.1 root 6654:
6655: @item EASY_DIV_EXPR
1.1.1.2 root 6656: An alias for a tree code that is the easiest kind of division to
6657: compile code for in the general case. It may be
6658: @code{TRUNC_DIV_EXPR}, @code{FLOOR_DIV_EXPR}, @code{CEIL_DIV_EXPR} or
6659: @code{ROUND_DIV_EXPR}. These four division operators differ in how
6660: they round the result to an integer. @code{EASY_DIV_EXPR} is used
6661: when it is permissible to use any of those kinds of division and the
6662: choice should be made on the basis of efficiency.@refill
6663:
6664: @item DEFAULT_SIGNED_CHAR
6665: An expression whose value is 1 or 0, according to whether the type
6666: @code{char} should be signed or unsigned by default. The user can
6667: always override this default with the options @samp{-fsigned-char}
6668: and @samp{-funsigned-char}.
6669:
6670: @item SCCS_DIRECTIVE
6671: Define this if the preprocessor should ignore @code{#sccs} directives
1.1.1.4 root 6672: and print no error message.
6673:
6674: @item IDENT_DIRECTIVE
6675: Define this if the preprocessor should ignore @code{#ident} directives
6676: and print no error message.
1.1 root 6677:
6678: @item MOVE_MAX
6679: The maximum number of bytes that a single instruction can move quickly
6680: from memory to memory.
6681:
1.1.1.2 root 6682: @item INT_TYPE_SIZE
6683: A C expression for the size in bits of the type @code{int} on the
6684: target machine.
6685:
6686: @item SLOW_BYTE_ACCESS
6687: Define this macro as a C expression which is nonzero if accessing less
6688: than a word of memory (i.e. a @code{char} or a @code{short}) is slow
6689: (requires more than one instruction).
6690:
1.1 root 6691: @item SLOW_ZERO_EXTEND
1.1.1.2 root 6692: Define this macro if zero-extension (of a @code{char} or @code{short}
6693: to an @code{int}) can be done faster if the destination is a register
6694: that is known to be zero.
6695:
6696: If you define this macro, you must have instruction patterns that
6697: recognize RTL structures like this:
6698:
6699: @example
6700: (set (strict-low-part (subreg:QI (reg:SI @dots{}) 0)) @dots{})
6701: @end example
6702:
6703: @noindent
6704: and likewise for @code{HImode}.
1.1 root 6705:
6706: @item SHIFT_COUNT_TRUNCATED
6707: Define this macro if shift instructions ignore all but the lowest few
6708: bits of the shift count. It implies that a sign-extend or zero-extend
6709: instruction for the shift count can be omitted.
6710:
1.1.1.2 root 6711: @item TRULY_NOOP_TRUNCATION (@var{outprec}, @var{inprec})
1.1 root 6712: A C expression which is nonzero if on this machine it is safe to
1.1.1.2 root 6713: ``convert'' an integer of @var{inprec} bits to one of @var{outprec}
6714: bits (where @var{outprec} is smaller than @var{inprec}) by merely
6715: operating on it as if it had only @var{outprec} bits.
1.1 root 6716:
6717: On many machines, this expression can be 1.
6718:
1.1.1.2 root 6719: @item NO_FUNCTION_CSE
6720: Define this macro if it is as good or better to call a constant
6721: function address than to call an address kept in a register.
6722:
6723: @item STORE_FLAG_VALUE
6724: A C expression for the value stored by a store-flag instruction
6725: (@code{s@var{cond}}) when the condition is true. This is usually 1 or
6726: -1; it is required to be an odd number.
6727:
6728: Do not define @code{STORE_FLAG_VALUE} if the machine has no store-flag
6729: instructions.
6730:
1.1 root 6731: @item Pmode
1.1.1.2 root 6732: An alias for the machine mode for pointers. Normally the definition
6733: can be
1.1 root 6734:
6735: @example
6736: #define Pmode SImode
6737: @end example
6738:
6739: @item FUNCTION_MODE
1.1.1.2 root 6740: An alias for the machine mode used for memory references to functions
6741: being called, in @samp{call} RTL expressions. On most machines this
6742: should be @code{QImode}.
1.1 root 6743:
6744: @item CONST_COST (@var{x}, @var{code})
1.1.1.2 root 6745: A part of a C @code{switch} statement that describes the relative
6746: costs of constant RTL expressions. It must contain @code{case} labels
6747: for expression codes @samp{const_int}, @samp{const}, @samp{symbol_ref}, @samp{label_ref}
6748: and @samp{const_double}. Each case must ultimately reach a
6749: @code{return} statement to return the relative cost of the use of that
1.1 root 6750: kind of constant value in an expression. The cost may depend on the
6751: precise value of the constant, which is available for examination in
6752: @var{x}.
6753:
1.1.1.2 root 6754: @var{code} is the expression code---redundant, since it can be
6755: obtained with @code{GET_CODE (@var{x})}.
6756:
6757: @item DOLLARS_IN_IDENTIFIERS
1.1.1.4 root 6758: Define this to be nonzero if the character @samp{$} should be allowed
6759: by default in identifier names.
1.1 root 6760: @end table
6761:
6762: @node Condition Code, Assembler Format, Misc, Machine Macros
6763: @section Condition Code Information
6764:
6765: The file @file{conditions.h} defines a variable @code{cc_status} to
6766: describe how the condition code was computed (in case the interpretation of
6767: the condition code depends on the instruction that it was set by). This
6768: variable contains the RTL expressions on which the condition code is
6769: currently based, and several standard flags.
6770:
6771: Sometimes additional machine-specific flags must be defined in the machine
6772: description header file. It can also add additional machine-specific
6773: information by defining @code{CC_STATUS_MDEP}.
6774:
6775: @table @code
6776: @item CC_STATUS_MDEP
1.1.1.2 root 6777: C code for a data type which is used for declaring the @code{mdep}
6778: component of @code{cc_status}. It defaults to @code{int}.
1.1 root 6779:
6780: @item CC_STATUS_MDEP_INIT
1.1.1.2 root 6781: A C expression for the initial value of the @code{mdep} field. It
6782: defaults to 0.
1.1 root 6783:
6784: @item NOTICE_UPDATE_CC (@var{exp})
6785: A C compound statement to set the components of @code{cc_status}
1.1.1.2 root 6786: appropriately for an insn whose body is @var{exp}. It is this macro's
6787: responsibility to recognize insns that set the condition code as a
6788: byproduct of other activity as well as those that explicitly set
6789: @code{(cc0)}.
6790:
6791: If there are insn that do not set the condition code but do alter
6792: other machine registers, this macro must check to see whether they
6793: invalidate the expressions that the condition code is recorded as
6794: reflecting. For example, on the 68000, insns that store in address
6795: registers do not set the condition code, which means that usually
6796: @code{NOTICE_UPDATE_CC} can leave @code{cc_status} unaltered for such
6797: insns. But suppose that the previous insn set the condition code
6798: based on location @samp{a4@@(102)} and the current insn stores a new
6799: value in @samp{a4}. Although the condition code is not changed by
6800: this, it will no longer be true that it reflects the contents of
6801: @samp{a4@@(102)}. Therefore, @code{NOTICE_UPDATE_CC} must alter
1.1 root 6802: @code{cc_status} in this case to say that nothing is known about the
6803: condition code value.
6804: @end table
6805:
6806: @node Assembler Format,, Condition Code, Machine Macros
6807: @section Output of Assembler Code
6808:
6809: @table @code
1.1.1.2 root 6810: @item ASM_SPEC
6811: A C string constant that tells the GNU CC driver program options to
6812: pass to the assembler. It can also specify how to translate options
6813: you give to GNU CC into options for GNU CC to pass to the assembler.
6814: See the file @file{tm-sun3.h} for an example of this.
6815:
6816: Do not define this macro if it does not need to do anything.
6817:
6818: @item LINK_SPEC
6819: A C string constant that tells the GNU CC driver program options to
6820: pass to the linker. It can also specify how to translate options you
6821: give to GNU CC into options for GNU CC to pass to the linker.
6822:
6823: Do not define this macro if it does not need to do anything.
6824:
1.1.1.4 root 6825: @item ASM_FILE_START (@var{stream})
6826: A C expression which outputs to the stdio stream @var{stream}
6827: some appropriate text to go at the start of an assembler file.
6828:
6829: Normally this macro is defined to output a line containing
6830: @samp{#NO_APP}, which is a comment that has no effect on most
6831: assemblers but tells the GNU assembler that it can save time by not
6832: checking for certain assembler constructs.
6833:
6834: On systems that use SDB, it is necessary to output certain commands;
6835: see @file{tm-attasm.h}.
1.1.1.2 root 6836:
6837: @item ASM_APP_ON
6838: A C string constant for text to be output before each @code{asm}
6839: statement or group of consecutive ones. Normally this is
6840: @code{"#APP"}, which is a comment that has no effect on most
6841: assemblers but tells the GNU assembler that it must check the lines
6842: that follow for all valid assembler constructs.
6843:
6844: @item ASM_APP_OFF
6845: A C string constant for text to be output after each @code{asm}
6846: statement or group of consecutive ones. Normally this is
6847: @code{"#NO_APP"}, which tells the GNU assembler to resume making the
6848: time-saving assumptions that are valid for ordinary compiler output.
6849:
1.1 root 6850: @item TEXT_SECTION_ASM_OP
6851: A C string constant for the assembler operation that should precede
6852: instructions and read-only data. Normally @code{".text"} is right.
6853:
6854: @item DATA_SECTION_ASM_OP
1.1.1.2 root 6855: A C string constant for the assembler operation to identify the
6856: following data as writable initialized data. Normally @code{".data"}
6857: is right.
1.1 root 6858:
6859: @item REGISTER_NAMES
1.1.1.2 root 6860: A C initializer containing the assembler's names for the machine
6861: registers, each one as a C string constant. This is what translates
6862: register numbers in the compiler into assembler language.
1.1 root 6863:
6864: @item DBX_REGISTER_NUMBER (@var{regno})
1.1.1.2 root 6865: A C expression that returns the DBX register number for the compiler
6866: register number @var{regno}. In simple cases, the value of this
6867: expression may be @var{regno} itself. But sometimes there are some
6868: registers that the compiler knows about and DBX does not, or vice
6869: versa. In such cases, some register may need to have one number in
6870: the compiler and another for DBX.
6871:
1.1.1.4 root 6872: @item DBX_DEBUGGING_INFO
6873: Define this macro if GNU CC should produce debugging output for DBX
6874: in response to the @samp{-g} option.
6875:
6876: @item SDB_DEBUGGING_INFO
6877: Define this macro if GNU CC should produce debugging output for SDB
6878: in response to the @samp{-g} option.
6879:
1.1.1.2 root 6880: @item DBX_NO_XREFS
6881: Define this macro if DBX on your system does not support the construct
6882: @samp{xs@var{tagname}}. On some systems, this construct is used to
6883: describe a forward reference to a structure named @var{tagname}.
6884: On other systems, this construct is not supported at all.
6885:
6886: @item DBX_CONTIN_LENGTH
6887: A symbol name in DBX-format debugging information is normally
6888: continued (split into two separate @code{.stabs} directives) when it
6889: exceeds a certain length (by default, 80 characters). On some
6890: operating systems, DBX requires this splitting; on others, splitting
6891: must not be done. You can inhibit splitting by defining this macro
6892: with the value zero. You can override the default splitting-length by
6893: defining this macro as an expression for the length you desire.
6894:
6895: @item DBX_CONTIN_CHAR
6896: Normally continuation is indicated by adding a @samp{\} character to
6897: the end of a @code{.stabs} string when a continuation follows. To use
6898: a different character instead, define this macro as a character
6899: constant for the character you want to use. Do not define this macro
6900: if backslash is correct for your system.
6901:
1.1.1.4 root 6902: @item ASM_OUTPUT_LABEL (@var{stream}, @var{name})
1.1.1.2 root 6903: A C statement (sans semicolon) to output to the stdio stream
1.1.1.4 root 6904: @var{stream} the assembler definition of a label named @var{name}. Use
6905: the expression @code{assemble_name (@var{stream}, @var{name})} to output
1.1.1.2 root 6906: the name itself; before and after that, output the additional
6907: assembler syntax for defining the name, and a newline.
6908:
1.1.1.4 root 6909: @item ASM_DECLARE_FUNCTION_NAME (@var{stream}, @var{name})
1.1.1.2 root 6910: A C statement (sans semicolon) to output to the stdio stream
1.1.1.4 root 6911: @var{stream} any text necessary for declaring the name of a function
1.1.1.2 root 6912: which is being defined. This macro is responsible for outputting
6913: the label definition (perhaps using @code{ASM_OUTPUT_LABEL}).
6914:
6915: If this macro is not defined, then the function name is defined in the
6916: usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}).
6917:
1.1.1.4 root 6918: @item ASM_GLOBALIZE_LABEL (@var{stream}, @var{name})
1.1.1.2 root 6919: A C statement (sans semicolon) to output to the stdio stream
1.1.1.4 root 6920: @var{stream} some commands that will make the label @var{name} global;
1.1.1.2 root 6921: that is, available for reference from other files. Use the expression
1.1.1.4 root 6922: @code{assemble_name (@var{stream}, @var{name})} to output the name
1.1.1.2 root 6923: itself; before and after that, output the additional assembler syntax
6924: for making that name global, and a newline.
6925:
1.1.1.4 root 6926: @item ASM_OUTPUT_EXTERNAL (@var{stream}, @var{name})
1.1.1.2 root 6927: A C statement (sans semicolon) to output to the stdio stream
1.1.1.4 root 6928: @var{stream} any text necessary for declaring the name of an external
1.1.1.2 root 6929: symbol which is referenced in this compilation but not defined.
6930:
6931: This macro need not be defined if it does not need to output anything.
6932: The GNU assembler and most Unix assemblers don't require anything.
6933:
1.1.1.4 root 6934: @item ASM_OUTPUT_LABELREF (@var{stream}, @var{name})
6935: A C statement to output to the stdio stream @var{stream} a reference in
1.1.1.2 root 6936: assembler syntax to a label named @var{name}. The character @samp{_}
6937: should be added to the front of the name, if that is customary on your
6938: operating system, as it is in most Berkeley Unix systems. This macro
6939: is used in @code{assemble_name}.
6940:
1.1.1.4 root 6941: @item ASM_OUTPUT_INTERNAL_LABEL (@var{stream}, @var{prefix}, @var{num})
6942: A C statement to output to the stdio stream @var{stream} a label whose
1.1.1.2 root 6943: name is made from the string @var{prefix} and the number @var{num}.
6944: These labels are used for internal purposes, and there is no reason
6945: for them to appear in the symbol table of the object file. On many
6946: systems, the letter @samp{L} at the beginning of a label has this
6947: effect. The usual definition of this macro is as follows:
6948:
6949: @example
1.1.1.4 root 6950: fprintf (@var{stream}, "L%s%d:\n", @var{prefix}, @var{num})
1.1.1.2 root 6951: @end example
6952:
1.1.1.4 root 6953: @item ASM_OUTPUT_CASE_LABEL (@var{stream}, @var{prefix}, @var{num}, @var{table})
1.1.1.2 root 6954: Define this if the label before a jump-table needs to be output
6955: specially. The first three arguments are the same as for
6956: @code{ASM_OUTPUT_INTERNAL_LABEL}; the fourth argument is the
6957: jump-table which follows (a @samp{jump_insn} containing an
6958: @samp{addr_vec} or @samp{addr_diff_vec}).
6959:
6960: This feature is used on system V to output a @code{swbeg} statement
6961: for the table.
6962:
6963: If this macro is not defined, these labels are output with
6964: @code{ASM_OUTPUT_INTERNAL_LABEL}.
6965:
1.1.1.4 root 6966: @item ASM_OUTPUT_CASE_END (@var{stream}, @var{num}, @var{table})
6967: Define this if something special must be output at the end of a jump-table.
6968: The definition should be a C statement to be executed after the assembler
6969: code for the table is written. It should write the appropriate code to
6970: stdio stream @var{stream}. The argument @var{table} is the jump-table
6971: insn, and @var{num} is the label-number of the preceding label.
6972:
6973: If this macro is not defined, nothing special is output at the end of
6974: the jump-table.
6975:
1.1.1.2 root 6976: @item ASM_FORMAT_PRIVATE_NAME (@var{outvar}, @var{name}, @var{number})
6977: A C expression to assign to @var{outvar} (which is a variable of type
6978: @code{char *}) a newly allocated string made from the string
6979: @var{name} and the number @var{number}, with some suitable punctuation
6980: added. Use @code{alloca} to get space for the string.
6981:
6982: This string will be used as the argument to @code{ASM_OUTPUT_LABELREF}
6983: to produce an assembler label for an internal static variable whose
6984: name is @var{name}. Therefore, the string must be such as to result
6985: in valid assembler code. The argument @var{number} is different each
6986: time this macro is executed; it prevents conflicts between
6987: similarly-named internal static variables in different scopes.
6988:
6989: Ideally this string should not be a valid C identifier, to prevent any
6990: conflict with the user's own symbols. Most assemblers allow periods
6991: or percent signs in assembler symbols; putting at least one of these
6992: between the name and the number will suffice.
6993:
1.1.1.4 root 6994: @item ASM_OUTPUT_ADDR_DIFF_ELT (@var{stream}, @var{value}, @var{rel})
1.1.1.2 root 6995: This macro should be provided on machines where the addresses
6996: in a dispatch table are relative to the table's own address.
6997:
6998: The definition should be a C statement to output to the stdio stream
1.1.1.4 root 6999: @var{stream} an assembler pseudo-instruction to generate a difference
1.1.1.2 root 7000: between two labels. @var{value} and @var{rel} are the numbers of two
7001: internal labels. The definitions of these labels are output using
7002: @code{ASM_OUTPUT_INTERNAL_LABEL}, and they must be printed in the same
7003: way here. For example,
7004:
7005: @example
1.1.1.4 root 7006: fprintf (@var{stream}, "\t.word L%d-L%d\n",
1.1.1.2 root 7007: @var{value}, @var{rel})
7008: @end example
7009:
1.1.1.4 root 7010: @item ASM_OUTPUT_ADDR_VEC_ELT (@var{stream}, @var{value})
1.1.1.2 root 7011: This macro should be provided on machines where the addresses
7012: in a dispatch table are absolute.
7013:
7014: The definition should be a C statement to output to the stdio stream
1.1.1.4 root 7015: @var{stream} an assembler pseudo-instruction to generate a reference to
1.1.1.2 root 7016: a label. @var{value} is the number of an internal label whose
7017: definition is output using @code{ASM_OUTPUT_INTERNAL_LABEL}.
7018: For example,
7019:
7020: @example
1.1.1.4 root 7021: fprintf (@var{stream}, "\t.word L%d\n", @var{value})
1.1.1.2 root 7022: @end example
1.1 root 7023:
1.1.1.4 root 7024: @item ASM_OUTPUT_DOUBLE (@var{stream}, @var{value})
7025: A C statement to output to the stdio stream @var{stream} an assembler
1.1 root 7026: instruction to assemble a @code{double} constant whose value is
1.1.1.2 root 7027: @var{value}. @var{value} will be a C expression of type
7028: @code{double}.
1.1 root 7029:
1.1.1.4 root 7030: @item ASM_OUTPUT_FLOAT (@var{stream}, @var{value})
7031: A C statement to output to the stdio stream @var{stream} an assembler
1.1.1.2 root 7032: instruction to assemble a @code{float} constant whose value is
7033: @var{value}. @var{value} will be a C expression of type @code{float}.
7034:
1.1.1.4 root 7035: @item ASM_OUTPUT_INT (@var{stream}, @var{exp})
7036: @itemx ASM_OUTPUT_SHORT (@var{stream}, @var{exp})
7037: @itemx ASM_OUTPUT_CHAR (@var{stream}, @var{exp})
7038: A C statement to output to the stdio stream @var{stream} an assembler
1.1.1.2 root 7039: instruction to assemble a @code{int}, @code{short} or @code{char}
7040: constant whose value is @var{value}. The argument @var{exp} will be
7041: an RTL expression which represents a constant value. Use
7042: @samp{output_addr_const (@var{exp})} to output this value as an
7043: assembler expression.@refill
7044:
1.1.1.4 root 7045: @item ASM_OUTPUT_BYTE (@var{stream}, @var{value})
7046: A C statement to output to the stdio stream @var{stream} an assembler
1.1.1.2 root 7047: instruction to assemble a single byte containing the number @var{value}.
7048:
1.1.1.4 root 7049: @item ASM_OUTPUT_ASCII (@var{stream}, @var{ptr}, @var{len})
7050: A C statement to output to the stdio stream @var{stream} an assembler
1.1.1.2 root 7051: instruction to assemble a string constant containing the @var{len}
7052: bytes at @var{ptr}. @var{ptr} will be a C expression of type
7053: @code{char *} and @var{len} a C expression of type @code{int}.
7054:
7055: If the assembler has a @code{.ascii} pseudo-op as found in the
7056: Berkeley Unix assembler, do not define the macro
7057: @code{ASM_OUTPUT_ASCII}.
1.1 root 7058:
1.1.1.4 root 7059: @item ASM_OUTPUT_SKIP (@var{stream}, @var{nbytes})
7060: A C statement to output to the stdio stream @var{stream} an assembler
1.1 root 7061: instruction to advance the location counter by @var{nbytes} bytes.
7062: @var{nbytes} will be a C expression of type @code{int}.
7063:
1.1.1.4 root 7064: @item ASM_OUTPUT_ALIGN (@var{stream}, @var{power})
7065: A C statement to output to the stdio stream @var{stream} an assembler
1.1 root 7066: instruction to advance the location counter to a multiple of 2 to the
7067: @var{power} bytes. @var{power} will be a C expression of type @code{int}.
7068:
1.1.1.4 root 7069: @item ASM_OUTPUT_COMMON (@var{stream}, @var{name}, @var{size})
1.1.1.2 root 7070: A C statement (sans semicolon) to output to the stdio stream
1.1.1.4 root 7071: @var{stream} the assembler definition of a common-label named @var{name}
1.1.1.2 root 7072: whose size is @var{size} bytes. Use the expression
1.1.1.4 root 7073: @code{assemble_name (@var{stream}, @var{name})} to output the name
1.1.1.2 root 7074: itself; before and after that, output the additional assembler syntax
7075: for defining the name, and a newline.
7076:
7077: This macro controls how the assembler definitions of uninitialized
7078: global variables are output.
7079:
1.1.1.4 root 7080: @item ASM_OUTPUT_LOCAL (@var{stream}, @var{name}, @var{size})
1.1.1.2 root 7081: A C statement (sans semicolon) to output to the stdio stream
1.1.1.4 root 7082: @var{stream} the assembler definition of a local-common-label named
1.1.1.2 root 7083: @var{name} whose size is @var{size} bytes. Use the expression
1.1.1.4 root 7084: @code{assemble_name (@var{stream}, @var{name})} to output the name
1.1.1.2 root 7085: itself; before and after that, output the additional assembler syntax
7086: for defining the name, and a newline.
7087:
7088: This macro controls how the assembler definitions of uninitialized
7089: static variables are output.
1.1 root 7090:
1.1.1.4 root 7091: @item ASM_OUTPUT_SOURCE_LINE (@var{stream}, @var{line})
7092: A C statment to output DBX or SDB debugging information before code
7093: for line number @var{line} of the current source file to the
7094: stdio stream @var{stream}.
7095:
7096: This macro need not be defined if the standard form of debugging
7097: information for the debugger in use is appropriate.
7098:
1.1 root 7099: @item TARGET_BELL
1.1.1.2 root 7100: A C constant expression for the integer value for escape sequence
7101: @samp{\a}.
1.1 root 7102:
7103: @item TARGET_BS
7104: @itemx TARGET_TAB
7105: @itemx TARGET_NEWLINE
7106: C constant expressions for the integer values for escape sequences
7107: @samp{\b}, @samp{\t} and @samp{\n}.
7108:
7109: @item TARGET_VT
7110: @itemx TARGET_FF
7111: @itemx TARGET_CR
7112: C constant expressions for the integer values for escape sequences
7113: @samp{\v}, @samp{\f} and @samp{\r}.
7114:
1.1.1.4 root 7115: @item ASM_OUTPUT_OPCODE (@var{stream}, @var{ptr})
1.1.1.2 root 7116: Define this macro if you are using an unusual assembler that
7117: requires different names for the machine instructions.
7118:
7119: The definition is a C statement or statements which output an
1.1.1.4 root 7120: assembler instruction opcode to the stdio stream @var{stream}. The
1.1.1.2 root 7121: macro-operand @var{ptr} is a variable of type @code{char *} which
7122: points to the opcode name in its ``internal'' form---the form that is
7123: written in the machine description. The definition should output the
1.1.1.4 root 7124: opcode name to @var{stream}, performing any translation you desire, and
1.1.1.2 root 7125: increment the variable @var{ptr} to point at the end of the opcode
7126: so that it will not be output twice.
7127:
7128: In fact, your macro definition may process less than the entire opcode
7129: name, or more than the opcode name; but if you want to process text
7130: that includes @samp{%}-sequences to substitute operands, you must take
7131: care of the substitution yourself. Just be sure to increment
7132: @var{ptr} over whatever text should not be output normally.
7133:
7134: If the macro definition does nothing, the instruction is output
7135: in the usual way.
7136:
1.1.1.4 root 7137: @item FINAL_PRESCAN_INSN (@var{insn}, @var{opvec}, @var{noperands})
7138: If defined, a C statement to be executed just prior to the output of
7139: assembler code for @var{insn}, to modify the extracted operands so
7140: they will be output differently.
7141:
7142: Here the argument @var{opvec} is the vector containing the operands
7143: extracted from @var{insn}, and @var{noperands} is the number of
7144: elements of the vector which contain meaningful data for this insn.
7145: The contents of this vector are what will be used to convert the insn
7146: template into assembler code, so you can change the assembler output
7147: by changing the contents of the vector.
7148:
7149: This macro is useful when various assembler syntaxes share a single
7150: file of instruction patterns; by defining this macro differently, you
7151: can cause a large class of instructions to be output differently (such
7152: as with rearranged operands). Naturally, variations in assembler
7153: syntax affecting individual insn patterns ought to be handled by
7154: writing conditional output routines in those patterns.
7155:
7156: If this macro is not defined, it is equivalent to a null statement.
7157:
7158: @item PRINT_OPERAND (@var{stream}, @var{x}, @var{code})
7159: A C compound statement to output to stdio stream @var{stream} the
1.1.1.2 root 7160: assembler syntax for an instruction operand @var{x}. @var{x} is an
7161: RTL expression.
7162:
7163: @var{code} is a value that can be used to specify one of several ways
7164: of printing the operand. It is used when identical operands must be
7165: printed differently depending on the context. @var{code} comes from
7166: the @samp{%} specification that was used to request printing of the
7167: operand. If the specification was just @samp{%@var{digit}} then
7168: @var{code} is 0; if the specification was @samp{%@var{ltr}
7169: @var{digit}} then @var{code} is the ASCII code for @var{ltr}.
7170:
7171: If @var{x} is a register, this macro should print the register's name.
7172: The names can be found in an array @code{reg_names} whose type is
7173: @code{char *[]}. @code{reg_names} is initialized from
7174: @code{REGISTER_NAMES}.
7175:
7176: When the machine description has a specification @samp{%@var{punct}}
7177: (a @samp{%} followed by a punctuation character), this macro is called
7178: with a null pointer for @var{x} and the punctuation character for
7179: @var{code}.
1.1 root 7180:
1.1.1.4 root 7181: @item PRINT_OPERAND_ADDRESS (@var{stream}, @var{x})
7182: A C compound statement to output to stdio stream @var{stream} the
1.1.1.2 root 7183: assembler syntax for an instruction operand that is a memory reference
7184: whose address is @var{x}. @var{x} is an RTL expression.
7185:
7186: @item ASM_OPEN_PAREN
7187: @itemx ASM_CLOSE_PAREN
7188: These macros are defined as C string constant, describing the syntax
7189: in the assembler for grouping arithmetic expressions. The following
7190: definitions are correct for most assemblers:
7191:
7192: @example
7193: #define ASM_OPEN_PAREN "("
7194: #define ASM_CLOSE_PAREN ")"
7195: @end example
7196: @end table
7197:
7198: @node Config,, Machine Macros, Top
7199: @chapter The Configuration File
7200:
7201: The configuration file @file{config-@var{machine}.h} contains macro
7202: definitions that describe the machine and system on which the compiler is
7203: running. Most of the values in it are actually the same on all machines
7204: that GNU CC runs on, so most all configuration files are identical. But
7205: there are some macros that vary:
7206:
7207: @table @code
7208: @item FAILURE_EXIT_CODE
7209: A C expression for the status code to be returned when the compiler
7210: exits after serious errors.
7211:
7212: @item SUCCESS_EXIT_CODE
7213: A C expression for the status code to be returned when the compiler
7214: exits without serious errors.
1.1 root 7215: @end table
7216:
7217: @contents
7218: @bye
This archive runs on limited infrastructure. Preserving old code on modern bandwidth. Automated agents are requested to crawl responsibly.